Techstrong TV – January 31, 2025

Hey everyone. We're all trying to figure out how this AI thing is gonna work in the enterprise. At the same time, we're faced with all sorts of new applications, new technologies, AI code generation, and yet another model coming to, uh, the, the world from China.

You're watching Textron Gang. Hey, everyone. We're coming to you live from San Jose.

Once again here for the Textron Gang. I'm Steven FoST, organizer of AI Field Day and the Tech Field Day events for the Futurum Group. And also, you might recognize me from Tuesday episodes of Textron Gang.

I am not Alan Shimmel, however, but I'm pretending to be be him. For these, uh, couple of episodes that we're recording live here. We are joined in San Jose by a great panel of independent technical folks who are, uh, participating, asking questions, discussing, uh, with our AI companies during AI Field Day.

You'll catch all of those things live streaming on Techstrong tv. But let's meet who we've invited to be on the panel with us today. So first up, uh, we've got, uh, a special guest.

Uh, once again, we've invited her back. Uh, Gina Rosenthal is one of my favorite, uh, tech thought leaders. I'm sorry to say that to you, Gina, but oh Lord, she is absolutely a tech thought leader.

Uh, joining us here on The Gang for the second day in a row. Whoop, Good morning. How are you doing this morning?

Well, it's okay. Uh, bad news outta dc Boy that, oh my gosh, that was some crazy, crazy stuff. That's, Yeah.

That's just hard to even think about. So, uh, also we've got another great, uh, gang member here, uh, John Willis. Great.

We'll take it. We'll take it right. Take 'em when they come.

Yeah. Right. Hey, yeah.

Great to be here with, um, you know, uh, the Tech Field Day has been great to be involved and doing, uh, tech King. Yep. Here live.

It's great, great stuff. And of course, uh, familiar face. Mitch Ashley.

Always good to be here. Good to be bumping elbows and talking to AI. And, you know, out here in the real world, I dunno if, if Silicon Valley's the real world, world, real world, yeah, yeah, yeah.

Sort of maybe kind of thing. Good, good to see everybody here on the gang. And welcome Gina.

Good to have you and time. Thank you. Yeah.

Yeah. You good to be back with you too, Steve. So, we, uh, yesterday, uh, had, uh, tech field day presentations by a number of different companies.

Um, the, the, the big ones, uh, that we saw that were sort of, uh, independent third party product presentations were, uh, from VMware and from member. And it was interesting during those sessions that we had, uh, both, both companies, at some point, some of the delegates internally were saying, Hey, these guys are trying to be the VMware of ai. Now, it's no surprise that VMware or Broadcom more specifically, it's no surprise that Broadcom's VMware team would want to be the VMware of AI because they wanna be the v VMware of everything.

I mean, remember, they wanted to be the VMware of storage. They wanted to be the VMware of networking. They want to be the VMware of cloud.

And so that makes sense. Me was an interesting one because, you know, previously I had seen, uh, their, their technology, uh, we had witnessed what they're doing. Uh, they have some very cool tech, but I didn't understand the scope of their ambitions until they presented yesterday.

And it became very clear that they also are trying to become the VMware of ai. In other words, they're trying to build a system that is scalable, that allows sharing of resources seamlessly, that maximizes your hardware investment. And the big thing that brought that home to me was in the first few minutes of their presentation, when they showed that chart showing that only, uh, you know, that, that, that the majority, uh, a third of enterprise AI installations are only used using about 15% of the compute resources they have available.

That was pretty mind bending. And it shows the opportunity for somebody to be the VMware of ai. 'cause you know, Gina, we were there at the beginning when VMware first started.

Mm-hmm. The reason VMware took off was because essentially the same stat. Most enterprises were using very little time of their resources, and there was an opportunity to do convergence.

So, uh, talk to us a little bit about what you see as an analogy. Well, first of all, I don't think VMware's trying to be the VMware of ai. I think they are the VMware of ai.

So they were, they've been working on, and this particularly when we're talking about merge, is, um, the GPUs, you know, you spend so much money on A GPU, you need it to be able to have the compute go fast enough to run all of these, um, jobs that, uh, building an AI platform will require. Um, so VMware's been working on that. You can, you can virtualize a VM or get to a, I mean A GPU or get to a GPU three different ways.

And that's been true for, since I was at VMware, so seven or eight years ago. So they have done it, but that's what they're looking at is how do we virtualize have a virtualized machine? How do I get a slice of A GPU?

And they also talked about wanting to make sure that, uh, the GPUs were being utilized across an organization and not hoarded, but shared by the vSphere Management system, which all of that does make sense. Um, and I, I think member is doing something very similar. They're looking at how do you provide, it was really cool.

How do you provide fractional access to A GPU and how can I put that in a management plane so that if I have three and you have six and you have 10, we have put the whole lot of those that the company owns into a, a platform that we can share the GPUs across. And then if it gets too busy on one, we can move 'em off pretty transparently, which is an awful lot like vMotion of GPUs. So, um, I, I think that it comes back to we we're looking at how do I make, instead of, uh, the compute on a CPU on a server shareable, how do I make those GPU shareables, those GPU shareable because they're so expensive to get so critical to ai, uh, workloads.

I think there's a spectrum of, of what they work on. And to me, VMware is how do you extend VMware into the world of ai, AI applications and AI hardware? Whereas Verge was, you know, in lieu of creating the next a MC show called GPU Hoarders, let's figure out a way to, to do GPU sharing and, and their approach.

They started with talking about the developer access through the IDE and being able to schedule that as part of your dev workflow. Um, and they were, and then they were adding all the things of snapshotting and, and restarting across different hardware, uh, which kinda led it to look like a bit of, of VMware. But my thought was, I don't know how member becomes the VMware of, of VMware of Gen ai.

'cause they're already gonna be it. That's, that's my, so they've gotta bifurcate some, by The way. Well, my point is, it's an apples and oranges conversation because GPU as a service and interest a service is a piece, and it might be the most interesting piece, giving all this constraints and things we've learned and we know about.

But the real problem is how do you supply AI in all of its spectrum, from containers, from virtualization mm-hmm. From, you know, things like Harbor and let's not forget networking you NSX, you know, uh, or you know, I'll go down the list, right? Go governance monitoring.

So when you talk about the, the Apple, which is Verge, which is a really interesting solution for a very interesting and hard problem. But there, when you start comparison Broadcom versus emh, you, you know, I I I pointed out that like the, I think you could use the analogy of do you run OpenShift or do you just piecemeal Kafka? You are an OpenShift.

You can get Kafka, you get policy, you get monitoring, you get all this stuff. And I'm, I'm, I, you know, I go back and forth depending on the client. Or do I do Kafka upstream?

Do I do, you know, do I do, uh, you know, Kubernetes upstream? Do I put in all the monitoring myself? You know, and I think Broadcom's gonna have a more compelling argument because the technical debt of doing all that yourself becomes a problem.

Well, and, but I don't think Broadcom is gonna attract new customers become, because they become the AI of VMware. I think it's the path for VMware customers. Exactly.

Yeah. Yeah. And it's an easier sell to walk in and say, Hey, do you really want to have a, a Kubernetes for this, a Kubernetes for that, a Kubernetes for that?

Or do you want to have a control plane? Mm-hmm. We kind of invented, We already got it.

Right. And The thing, the thing about that is though, that's what you have to buy. So if I'm a, if I'm a broad current Broadcom customer, I know that I can only buy the whole kit and caboodle.

Right. And it's expensive, right? So yeah, it, it might be the best in class and have everything all there, but you're still gonna have to pay for that if, even if all I want is to have the ability to, uh, democratize the GPUs in an organization.

For me, what I thought was interesting, I kept thinking it sounds, when they were talking about, um, when Member Ver was talking about sharing the GPUs, I kept thinking about, this just sounds like SRDF to me back from the EMC days. And then having a conversation with the CAOI find out, oh, you're a storage guy, so you're approaching this as, how do you know we've got this data sitting over here, we gotta get it in really fast over here. I've got this equipment that's slowing me down because I can't get to all the equipment.

How do we solve that problem? So I think that's an interesting problem. And I think if, you know, you're talking about they've got that they're sitting on top of a layer of Kubernetes.

So if I've infrastructure as code infrastructure is coded, all the things that I have, I should be able to build those pipelines and get it out there. I mean, that's another thing, right? When you talk about, like, the things they don't, didn't show, uh, Broadcom yesterday is when I asked like, what script?

Well, there's years of Terraform in there. Yeah. Like, that's not gonna show.

I mean, you know, they're, they're based on the questions Verge, again, I'm rooting for 'em. They were very immature about what it was gonna take to manage Kubernetes. And, and that's gonna include a lot of stuff that historically has run at scale.

The only thing I, I mentioned to, I, I talked to the, the CEO last night too, and I, you, I'm gonna give you, I had, you know, three margaritas. So I'm like, yeah, like, I think you need to go open source. You know, you need to become the Lang chain.

If this is gonna something you're gonna solve, you need to get in the middle of everything. 'cause you're gonna have such an uphill bo battle dealing with the Broadcom messaging and whoever else is gonna come out with this, this sort of like, full solution. You know, red Hat obviously is going to have some, they do it in Instruct Lab and stuff like that.

So, you know, in order for them to get, like, become the lang chain of, of this new stack, you know, so I, I don't know if you, you Know, well, it's like sitting between, you don't need to be VMware. You already have that. Mm-hmm.

You, you're running on Kubernetes. I'm not sure they know what they've got there yet. Right.

Right. All the capabilities that, that, uh, has as well as what they can do, but they want to be the stack on top of it. Right.

There's a lot of companies pushing for the, how do I get AI into production? There's the neural magics that, uh, right. Red Hat bought, right.

Very much down that lane, Steven Chair. And these guys could be, uh, on that same path. And I think there are other, I mean, it's, I think it's gonna be a very competitive, because Kubernetes doesn't step in and solve all of it, right?

No. It's just a workload Control plane, and it is just a control plane. Exactly.

And, and it may not be a great control plane for ai. You know, that was one of the things that came up in the discussion was that, uh, you know, many of the aspects of Kubernetes that makes it compelling for, uh, webscale modern applications makes it not all that compelling to run, uh, machine learning training and inferencing and so on. Especially The training.

Yeah. Yeah. And so it's gonna be interesting to see where that goes.

But I think that the reason for me, I, I think I'm basically just exposing my n tech nerd, hardware nerd background, because, you know, when I look at what they're doing with the fractional GPUs and GPU sharing and splitting and optimizing infrastructure utilization, it does, it reminds me so much of the messages that we got from classic VMware way back in the day, really, you've got this incredible investment in hardware. It's being underutilized, you've gotta utilize it. But I think you're right.

I, I'm gonna say that, you know, number one, we have the VMware of ai. It's called VMware. Yeah.

Uh, number two, uh, MERG is a very different solution. And, you know, more of a, more of a, a technical solution that may end up finding its way in a completely different way than, you know, becoming some sort of, uh, you know, hypervisor for training to Your, to your point, I mean, they're starting at, obviously at different places, but they're talking in different audiences, right? You're VMware's talking to the running very large infrastructure.

Mm-hmm. So they're solving the AI of VMware from that perspective. Members is gonna be talking about a AI development and model training, AI engineering teams trying to do work to get things into production.

They're gonna solve different problems. I, than I think VMware will ever get to. I Don't know if I fully agree with you.

I know we talked about it earlier, because if you, they did have, they did focus on the assist admin, the DevOps, right? But, but that was to create the developer experience, right? So I think there is a path of like, development experience, and they did talk about, you know, sort of GPU management, and there was a difference between how they did that.

So, and, and the other thing, I'm, again, I'm, I'm not like well-versed in, you know, CUDA or Nvidia APIs all, but like, it sounded like Verge is really just using most of what, um, NVIDIA provides. Mm-hmm. Right?

There's no, like se you have to work with their management tools or APIs. And, and so, like, to me, that just says broad count will be able to catch them as quick, you know, if it's all exposed and they're just using, what's the IP that comes out in nvidia? I'm not sure.

I think there might be a race problem. That's why I think they're gonna have to find what it is they're gonna pivot to. That was the, one of the que they really have to be clear about what exactly are they doing.

Uh, and, and it's A very interesting product marketing problem from a messaging perspective, because both companies, the tools are, are, are used by operations people. And operations people don't really, you know, necessarily, we, none of us really have the vocabulary and the understanding for how an AI workload works. It's, it's just a workload.

So you should be able to manage it and make every, all of the hardware performant and do everything that developers want to do. And that's the, the goal of now a platform engineer, which is assist admin in my view, but, you know, help me, I'm wrong. But to get the platform engineers to a place where they have a self-service platform for the developers.

The developers and that team, the product owners are running that show for how we want to share our GPUs or how we want to do things. So you may have this IT person that's connected to the IT department that's specifically trying to figure out how do I get them to stop fighting over the GPUs? How do I make this so I can tell when the server, the GPU is attached to is gonna go down and I can provide all the security and data protection and fault tolerance and everything that I need to provide to keep a business running.

But they've got a really weird messaging problem where the people that are gonna be using it and need to understand how the, the developers work, so they, they can serve that team, have no say in the buying at all. And that's, that's a really hard problem. So you have to have, that's why they both led with this.

We know you got a bunch of GPUs sitting out there underutilized. That's a lot of money you're wasting. Let's fix it.

Because that's a problem everybody can Understand. And, and I'll tell you this. So I'm glad you brought that up because one, I, I worked, I was early in the docker, right?

And Pivotal killed us because you know who they sold to the infrastructure people. Mm-hmm. Right?

And the infrastructure people who are by, that's gonna be Broadcom's, they're walking in. We were saying, well, should they be more demo? The, the most dangerous thing you can do is take the Docker out where you tried to solve you.

You believed that if you could get billions of developers using your tool, you'd win. And it, that, that theory did not play out. I mean, Docker won, but Docker as a corporation lost.

And, and, and it really was because the competitors of Docker were focusing, you know, pivotal. Um, you know, then, um, you know, Mesosphere, like they, they were coming in at the infrastructure level. And Broadcom's walking into your point, they already have the infrastructure and you have to win the infrastructure.

They have it. Yeah. And they already own it.

And Broadcom, that's the thing. I think that's important here. Broadcom, when bought VMware, uh, Broadcom already Yeah.

Was a leader in infrastructure. Now they're even more a leader. Yeah.

Yeah. I mean, Broadcom with VMware is even stronger than VMware ever was in that market. And you're right.

I think that that's where a lot of the money is. Yeah. That's where a lot of the buyers are.

And, um, you know, and, and so I, again, I think that shows that these are not like competitive products. These are two completely different products, even though the messaging is similar. Well, we do have to move on, uh, to our next segment.

And in our next segment, we're gonna talk more about developers, uh, keep watching Discover Techron Group, the epicenter of tech innovation. We are your go-to for reaching IT leaders and practitioners worldwide. Our secret impactful content that sparks awareness, engagement, and top quality leads with us.

You'll access editorial websites, streaming videos, virtual events, custom content analyst research, and more. Join our satisfied clients. Let's revolutionize your tech journey.

Contact us today and tell your story to the world in the most powerful way with Textron Group. Welcome back to the Textron Gang. So we are going to turn the page here from, uh, infrastructure to developers, and we're gonna talk about one of the primary applications of ai, which is coding.

Now, I think that there's this thought that somehow a chatbot can just spit out a functional application with no real input. And we don't need all those developers anymore. But I'm not sure that's really the case.

So John, uh, talk to us a little bit about the landscape of AI coding tools. Yeah, no, I mean, you know, there was the chat GPT moment, then there was the co-pilot moment, which sort of were really close to each other. And they, they had similar effect on different groups, like the developers.

Immediately, a lot of people like just is nonsense. This is terrible. You can't do it.

It'll generate terrible code. And, and that was all baloney because I, you know, I talked to banks that have 18,000 Java developers, and I'm like, what percentage of them actually create great code? You know?

And like, so let, let's get off of this, that the AI agents create bad code, right? Like, it actually creates better code than most of the coders in your organization. So then, like Copilot was interesting, but then what came out, if you've heard Cursor, cursor sort of blew everybody's mind because they, somebody got their 9-year-old kid to write an application with it, which again, was more marketing and nonsense, but it sort of changed everything about Copilot is just like, gimme Code Cursor was like intelligence in the IDE.

And right now there's like three or four products that are really interesting, cursor ada, but it, you know, and now it's getting confusion. Like, what do we, you know, what do I use John? Right?

And I, I've been working on this like blog article that I'm trying to write. Like, the truth is you're gonna use more than one, and it's gonna be based on the test that you do. Like, if I'm just wanted to prototype an idea, I might just do that in code.

If I'm gonna refactor code, I'm going to use probably Cursor or something called ADA or client, right? So, and you're probably gonna use, you're gonna come up with your sort of best two or three. And then you start understanding like, cost.

And then I read something interesting that like 80% of whatever coding tool you use is really comes down to the model you use. 20% is the interface, right? So now you gotta really think about the coding tool and the model where 80% of what you're going to accomplish is based on the model you use and some model combination.

So there's some really interesting, I think as cut as executives and team leads try to decide, should we just pick one product and everybody has to use it? Do we allow them to have freedom? I think, you know, and then there's the sort of scale of the coder.

There's junior coders. Like for me, I'm not an industrial coder that codes every day, eight hours. I like Eighter, friends of mine who are industrial coders, like cursor it, it appears, it appeals to them better.

So the IDE itself, the tool, um, so any, I, I think there's just like token costs, use cases multimodal. And one last thing I'll say is context is interesting. Some of the tools actually allow you to pull the full context.

You know, what happens when you go into copilot? The context is just code. And that's not really how a software engineer builds an application.

No. They have documentation, they have Terraform scripts, they have all these non functionals. And if the AI tool is not aware of those things, you're sort of, uh, collaboration with the tool is limited.

So I think what we're finding is some of the, these tools that really allow you to expand the context reads in a repo and understands not just, you know, runs into how it was built. What are the dependencies, you know, what, you know, what's the documentation? Even Jira tickets, right?

That kind of stuff. So some of these ones that I'll just, you know, ADA, that ADA seems to do a really good job, and there's a new one called Windsurfer. And then as you pointed out earlier, in all that craziness Monday of, uh, deeps seek, um, bike dance, put out something called Trey.

And I haven't, I'm for obvious reasons, I haven't tried it yet. The same reason I didn't use the API over at Deep Se, but, well, I think you did a really good job of kinda laying out the, sort of the emergence of all the different tools, right? That are built on what kinds of models.

Um, and a couple things that you illuminated. One is it's context as in what's the whole code bank? What are the other things that developers do?

Right? What, what is that whole thought all mind process of, of managing and delivering software? Not writing a piece of code.

It's like the equivalent of what model generates the best code. It's like, which model generates the best TV script versus a blog post, right? But you need really the full context of it.

And a lot of what drive drives it is graph technology that interconnects, here's your, here's your Jira tickets, here's your open items that you're supposed to, here's the, the technical debt backlog. Here's five ways that we do have implemented this particular coding pattern in our software. What's the best application of it for this situation?

That's what a developer looks at, right? Yeah. What do I know about how we do?

So, so we're, we're still at the experimentation stage, which is a lot of the choose the best tool for whatever, and still focused on the point of generating code. I think the next step beyond that is really taking it to what's a real workflow Look like? And it isn't black or white.

It isn't like cursive versus this versus code. Yeah. It's like, is it a quick fix?

Is it a refactoring, which is a big job. Like the different tools will have better workflow for the type of thing you do. 'cause developer might be somebody who's just fixing to do commits and development might be refactoring.

Mm-hmm. Like, you know, 40,000 lines of code. Those are different type of things that require the tools react differently for those.

So to your point, my best advice is do the bakeoffs. There's some great video bakeoffs out there. Literally try to just take an application, take a source code base that you know really well, and try it in different scenarios.

Quick fix, refactoring, uh, from scratch, you know, new componentry, new features. And then you'll get a feel for which of these tools actually are gonna work Kind of beat, take an engineering approach to it, right? Yeah.

Like who, go figure. Yeah. Very much so engineering.

Yeah. Well, and there's others as tab nines, there's also companies who are doing a lot of work around analyzing code bases, like for app modernization, whether it be mainframes Yeah. What job or whatever it is, just telling you what this code does.

What are the, like what, what it's, what is it solving? What Functions? Well, the good is, is a lot of those have all that built in app.

Yeah. And that's combining those things. Exactly.

Right. So it's the, the point of the article that we kinda surfaced that started this conversation is develop, well, how much time do they spend writing code? But also, how much time are you gonna spend understanding and fixing code that's generated for you?

Right. Or just, you know, you know, the developer days of I'll just do it myself. It's faster.

Right? Well, when that goes over past that productivity hub, no, I'll just take what they gave me and I'll fix it when I need to fix it. Right.

I'll use the tool to help me understand what it's doing. You know, that sounds terrifying to an op person because work worked in dev is a thing. Right?

Hey, it would've worked in worked In production. Yeah. But I mean, But no.

And, and, but That doesn't change, right? No, it doesn't sound like it's gonna change Change at all. We, We, I Think it just happens.

Stack overflow Cut, cut and copy and paste to now, you know, so, I mean, 'cause look. Yeah. Give me, yeah.

Like, I could get wrong. Yeah. And, and I think that one more thing that I'll throw in here, and I think probably Gina's thinking about this too, because I know, I know what your, uh, your concerns are about a lot of these things is the privacy aspect.

So when you're talking about, you know, oh yeah, you know, maybe you put your code base in this one and try it out. Put your code base in this other one and try it out. I'm like, wait, whoa, you are gonna do what with your source code, you're gonna put it into, you know, your proprietary enterprise source code and all of your commits and all of your bug reports and, and, and everything.

You're gonna put that into an AI engine that is, uh, you know, and so Cursor, for example, uh, claims, you know that they're soc two compliant and that they keep everything local. I like that. Um, not endorsing it by any means, but I'm just saying, you know, I'm glad that they surfaced that right away by the, but, but that's my concern is, you know, you put your code into these things.

How do you know where your code is going? You Don't. Well, but when you enable, uh, Google Gemini on your Google Workspace account is access to everything.

A lot of that's happening with it's, it's happening all over the place, but it doesn't, it's a legitimate concern. But, but the thing about it is people don't know. So this is going back to like, the ops people don't understand the architecture.

So we're having a hard time figuring out how do you, how do you manage everything? But I think for all of us, like what if you, if you put something into Google or Microsoft and you don't have a paid account, at least with a paid account, you've got at least the, I dunno if it's a facade, but you got the, they tell you that they're not gonna do anything with your, your per personal data. If you go to some of these other new ones that are popping up, all bets are off.

And I think people need to educate themselves about that. It's so important. Gene, I don't know if this will, this will help, but I can imagine a day not very far down the road that because we understand your code base, AI does, maybe it generates some of the codes.

Some of the not and ops person can say, why is Kubernetes doing this? Absolutely. And it isn't gonna say, well, Kubernetes has these three features, and here's how they work.

It's gonna say, here's what your code, our code is doing, and here's three things you might do. Right. That's, and I think that the, that's, can you imagine that it, it worked in dev, it's not working in ops because of all the, all the other things we have to do in ops to lock everything down.

And I can just ask a question and it tells me. Yeah, that's amazing. And, and, and again, the models itself, you know, like, I mean, and I, I wouldn't trust like NSA code to this, but like OpenAI, they don't where they state that we not trained on your data.

Right. Like, I still think there's dragons out there of putting anything in context, but at least there's model providers that clearly state we do not train. They know it's an issue.

Yeah. But you have to pay for it. And in my experience working with devs, they, I mean, I, I used to work one place that I worked, um, we had, it was astrophysicists, postdocs, and they had to, um, get on our network and we'd get a ticket if they tried to plug in.

And we'd go and help 'em out, help them, you know, fill out all the protective forms they needed to have. One time I went to look, and the dude, I swear to goodness, had like an all cardboard box with a towel in it that like, you put lost kittens in, and he had breadboarded his Linux machine together. And I was like, this was way before like cell phones.

So I don't have a picture, but I just turned around and went to get my boss. Like, I'm not even like An Academic research Environment To open things Up. The good news is, a lot of the large corporations that I've been working with are really setting standards on things like I, they have to, what, here's the only copilot you can use.

And it's already, you know, so, and you know, again, do you, I'm Sorry, I'm laughing at that. Just because any, I mean, setting corporate standards for anything is not always that effective, especially when it's happy as easy as going to a website. I, I agree.

But they have to do that. Otherwise it's just a free for all. You know what show?

Oh, I, I'm not saying they shouldn't, I'm just saying that it's not gonna work. Well, It works in the sense that as we get, you know, there's another whole subject, but, but the whole sort of auditing and, and I think what's going on in, you know, what we wrote about Investments Unlimited, that what's going on there about making sort of DevOps auditing, if you will paraphrase. Um, it, it's putting more light on the things that you're not doing as organization.

We're informing internal auditors way better now than we were pre five years ago. And that puts a flashlight on use. 'cause you have to document the chain, the supply chain of what you did.

You know, the how did you get from your code to deployment to all that stuff. If you're sort of creating attestations around that Fantastic book, by the way Yeah. Download that, that allows for the organization to start finding those shadowy.

We have to talk to the individual, everybody, because right now there's so much hype around ais and models and do this and do that. And it's exciting and it's, oh my God, I wanna go try it. And I think there needs to be a little bit of education of like, I know you wanna try it, but really, really think about do what happen if you connect to this api.

Mm-hmm. What will happen if you do anything with data you really care about, or that your boss might really care about. Yep.

Yeah. I, I, The hype, Is it somebody said, uh, I think it was John, uh, that this was the greatest data exfiltration Oh, Monday, um, in, in history. Because by dance or, um, not by dance, because deep Seat got so much attention that, um, you know, probably literally millions of people register, dumped all sorts of sensitive information into it immediately.

And that was that. Yeah. TikTok went, why didn't we think of That?

That's a good one. So we gotta move on. Thank you very much.

Uh, this is a conversation. I know that's gonna continue on the Textron Gang. Uh, up next, uh, we're gonna look at a new model, uh, coming in from China called Quinn.

Stay tuned. com is the leading resource for news analysis and education on challenges facing the cybersecurity industry. com covers all aspects of cybersecurity, including data security, DevSecOps, cloud security, application security, network security, security threats, and more.

com has the largest selection of security content featuring breaking news, blog posts, podcasts, and more. com to learn more. com.

Home of Security Bloggers Network. Welcome back to the Textron Gang. We are at AI Field Day, and, uh, if you've noticed, every one of our conversations has been about ai, but that's not all that unusual.

I think everything on every episode of Textron Gang for the last couple of months has been ai. Yeah. Uh, let's talk a little bit about another new, uh, model coming outta China that claims to be even better than deep seek.

5. Max says that it outperforms deep seeks R one, and that it uses, um, low amounts of energy and all this kind of stuff. Essentially, it's the same story again from, uh, Alibaba, which is a huge, uh, Chinese cloud company.

Uh, who wants to take this to start? I, I'll go just quickly. I mean, Quinn has been around for a little bit.

Um, so, uh, and there's been versions of it. Um, you know, Alibaba's real, I mean, you know, Alibaba Cloud is real. It has been real.

It's been, you know, there, there have been a number of retail companies that used Alibaba Cloud because you couldn't use Amazon and China. So if you were Starbucks, you literally had to bifurcate your cloud infrastructure between sort of Amazon and Alibaba. So Alibaba has been, you know, like sort of vetted to the quote that any company based out China can be.

Um, the Quinn models are, are interesting. If you follow Ruben Cohen, I'll try to get some show notes for y'all. Uh, he's done investigations going back a couple of months now on the, the, all the China models, including Deep Sea, are very restrictive.

Like, so they create inference guardrails on things like, you can't ask about Tandem Square, you, and the, that, you know, that's just the obvious. The unobvious is when you put those constraints in an inference model, the algorithm type models that you do, it actually creates more B for Jailbreaks. And Ruben was able to show some really good examples of how you can use inference jail breaks to get it to ask questions that the model itself is designed not to answer.

You know? So, um, so again, I think this is a, this will be a problem that will bubble up on a lot of the, the models that come outta China. The fact that they create restrictions in these sort of algorithms themselves, or the inference actually makes them more vulnerable to Jailbreaks.

I wonder if it, if it sort of sets the race, the space race to get to general AI a bit to the side, right? We're all focused on kind of what the next model is. That's who's gonna get to the singularity first.

And I was making this point the other day. I think it was on Monday or Tuesdays show, talking about, yeah. But there's also good enough ai, not everybody needs, you know, the singularity to solve every problem.

NGI and a I are false flags. Yeah. I they can, you know, the, the, the, it it takes us, it's a distraction.

It's a distraction from absolutely what do you need for ai? Let's stop worrying whether it's gonna take over humankind or termin. And it's, and it's not a single Point.

You're scientist. Let's be engineers. Yeah.

Yeah. Work with what we got. Thank, It's not a single point.

Isn't, isn't general AI gonna be Well, like what? Q like Q General, it steal that. My general AI is smarter than your, anyway.

I think that's sort of a Yes. That's part of the race. The money's made.

'cause nobody's gonna be able to afford to run it, right? I mean, that's gonna take the, the highest compute of anything, everybody. What what's interesting is the application of AI and how accessible it is to people.

And I think that's why TikTok being jealous of, of, uh, a deep seek. I called it deepfake the other day by accident. Oops.

Uh, I guess Freudian in some ways, but T tiktoks got its own stuff going on. It's the cult. They're after the culture.

They're not after the They're they are, but they wasn't quite, they're also by dance though, by the way. It's all coming from by dance. So, yeah.

So, So for me, I think if I look at it from a product marketing, marketing perspective, it's not an accident that these two models were, um, released one right after the other. I don't believe, especially a few days after the us um, government had a huge announcement about, uh, investing Stargate tons of money into Stargate. And this is what we're gonna do as an, as our country to, to, you know, get up on AI and, and get all of this done in, in this side of the world.

Um, it's also, they announced yesterday was Chinese New Year. So that's a great big celebration day. I think that's a thing too.

Um, and I, I kind of find it, it, it's all around the hype, right? Which is what marketing does. We drive the hype to get everything excited, everybody excited about it.

But the, the market is not the same in China as it is in the West. It is controlled by the government. Everything that happens is orchestrated by the government.

So, and, and allowed or not allowed by the government. Mm-hmm. Mm-hmm.

So, very interesting that, you know, we have to start, we have to look at it. China is a, a foreign adversary to America. And we have to look at it from that perspective, because we are in America.

We're not in China, and we are a global universe. And we do want to cooperate and, and be colleagues with all of our people, all of our friends, um, from other countries at the same time. They're not playing by, they're not playing the same playbook that our companies play by.

And the playbook has changed too, right? Just as an author and, and like, you know, um, five years ago, if I was offered to speak in China, I'd go like that. 'cause you sell crazy amount of books.

When you, if you're an author in China, I won't go near that place right now, if you read about people being detained so that the climate of even the global economy within the last three or four years has changed these as a person who speaks, I know we all speak around the world, right? Right. Like, I wouldn't go, I would, there's no amount of money I would go into China, right?

Well, And I'm not, and I'm, and I'm not saying that as like a down thing. I'm just saying, you know, let's look at this from, if I'm looking from a marketing perspective, what are they trying to do? What are hype are they trying to achieve?

Because we know that there are definitely some constraints as far as privacy goes with those models and sharing information. But, so they wanna get it hyped up, get everybody excited about the next model. And they have done, let's be honest, some really cool engineering work.

That part's amazing, but why is it so hyped? Why is it so why did they pick the timing? And what does the, is what does the Chinese government have to, to win from that?

And, and also, I think we should point out too, as as was said at the beginning here, uh, deep seek didn't come out of the blue. Uh, this was, you know, just the latest iteration from a company that it, you, you know, you would've seen this coming if you had been watching closely. Uh, the same with Quinn, uh, you know, as we just heard, you know, there's been other Quinn models.

Uh, this is just the newest one. Okay. Um, it seems pretty good.

And, and as we talked about yesterday, uh, when talking about deep seek as well, uh, in many cases, the us uh, restrictions and controls on, uh, on AI components, forced them to build these models in a more efficient manner. So just like deep seeq, you know, the Quinn is a mixture of experts model with, uh, that uses a lot of small models kind of teaming up together. Um, it makes a lot of sense.

You know, this is the, this is the microcomputer versus the mainframe in the, in AI sense. And, and so it's not really a surprise that this is, that this thing came, and it won't be a surprise next tomorrow or next week when another model comes out claiming to beat, uh, deep seek or chat GPT or, or whichever the benchmark is. But ultimately what we're talking about here is basically who's got the biggest, baddest, fastest, strongest, when what we really should be talking about, I think, is, uh, the application side of things, which is what we were talking about.

Yeah. Earlier on this discussion. Well, Elevated a level.

You mentioned the marketing of it. It, it's, it's geopolitical marketing of it, right? Because, because China's the way it is, then it creates all kinds of conspiracy theory or reality, whichever you choose to believe is, is this the Star Wars moment for China of setting this big thing that now we have to go, go chase them after, kinda like Reagan did with Star Wars back in the eighties, right?

Or is this something legitimate? And I think, by the way, self, self shameless plug, either today or yesterday, um, Futurum research came out with a paper, our kind of position thinking about it, different parts of Oh, cool. The deep seek announcement Daniel Newman contributed than I did as a bunch of other analysts.

So go check that out. com. Well, we do have to wrap up now.

Uh, thank you for Joining us. I, Wait, you got, Had you had One point, Okay, hit hit. There's no doubt China is an adversarial nation.

But, but yet Monday and Tuesday, we heard a lot of, um, well, China, you know, oh, there's America and China. Well, you know, the stuff that came outta France Mystro. Yeah.

So there wasn't so in on the geo, geo computer science spectrum, this is just d different countries that are just showing that they have ip, they have knowledge. But I can't exclude the fact that like, I'd be more trustworthy of Mytral from France than I would from Quentin. When, so when did minstrel come out?

That Was a year or two years Ago. Okay. That's what I'm saying.

I think it's very interesting. No, No timing. Oh, I totally timing.

I totally agree with you. Timing's everything. It's Interesting.

Yeah. I totally agree with the timing point. Anyway, Anyway, I was saying awe, because I was sad.

We were, it was over. Well then we will, we will have to, I'll, I'll, uh, put the screws on Alan to invite you back. Yeah.

Woo. And, uh, I get to meet Alan. That'll be awesome.

That would be amazing. Um, so thank you very much for watching. Uh, tune in, uh, uh, Textron TV for all of the AI Field Day sessions.

Uh, we've got some great presentations, discussions. We're gonna do a round table discussion. Uh, we've got the, uh, weekly Tech Field day podcast streaming on Textron tv every Tuesday.

Uh, and of course, uh, check out the Textron Gang every day, every weekday. Uh, you'll find, um, at least three of us, and I'm gonna guess four of us, uh, back on the Textron Gang in the, pretty soon in the coming weeks. And, uh, we'll see you there.

So thanks for watching. And Alan, um, uh, come on back. You're our only hope to rescue us from, uh, from my hosting.

Uh, take care. This is Techstrong tv. Hey everyone, we're back Tech strong tv.

We've got another CEO to talk to. Let me introduce you all to Tom Fin. Fin link.

ai. ai, but hopefully after today you'll know who they are and what they do. Let's welcome Tom here, though, to text on tv.

Hey, Tom, how are you? Thanks for joining us. I'm doing pretty well.

Yeah, how are you? Very well, thank you. So Tom, we're gonna talk a lot about Conifers, but before I do, let's talk about you, you know, you're the CEO here.

Give us kind of, how did you wind up being the CEO at Conifers? Oh, yeah, that's a very interesting story. So, you know, started my career, uh, back there in Tel Aviv as part of the intelligence community over there, doing a lot of very interesting, exciting things we can talk about.

Um, then after my military service, I joined a little company called VMware. Um, back in the days, uh, when we did public clouds, private cloud, and a lot of interesting things in the data in the data center, um, I had a pleasure to join the Israeli office back then in Tel Aviv and work on the European market. Uh, but as our offerings took off, um, I was asked to move to California, to Palo Alto, to headquarters, um, and lead the, um, portion of a product team of the cloud management division.

Uh, so this is where I first exposed to data science, and we did monitoring and data science, uh, for the data center about 12 years ago. It was remarkable. It was really interesting.

Uh, we got a couple PhDs helping us, and, you know, we needed to have, you know, mountains of servers in the server X in the data center just to run some simple predictive analytics, um, and machine learning. Um, after that, uh, did a lot of work on move to the startup side of the house and ran product and engineering, uh, for a sustainability startup. When we did optimize, uh, consumptions and generation of power plants in the United States based on predictive analytics, was very, very interesting.

Um, and then I moved to insights, uh, doing, um, threat intelligence, uh, great company. Uh, I moved to the dark side, moved from the product and engineering side of the house into the sales side of the house, and I was the chief customer officer over there, uh, six and a half years. Great journey 2021.

We were sold to Rapid seven on a 350 transaction. Um, and then I joined the detection and response, um, practice at Rapid seven, um, as leading product management over there. Um, this is where our exposed to, again, MDR services, the soc, um, and a lot of the different phase that we are doing today at Conifers.

Uh, great run over there. Um, wonder more experience meeting a lot of great people and customers. Um, and this is what inspired me, see the challenges, see how you provide soak services in large scale, uh, and see what's working and not working, gave me the opportunity to kind of understand what, what we think, what I think what I believe should be the next stage in detection and response and security operations.

Um, and inspired me to kind of, uh, open conifers and start with Conifers. Excellent. What a journey.

What a great journey, man. All right. Oli Oli in the tech world.

Do you hear, you know, these kinds of fantastic, fantastic, uh, career paths. Um, what made you found or get involved with Conifers? What, what, what was the passion behind that?

So, when I was at Rapid seven, um, I was implementing a data science project within the Fred Intelligence, um, product line that we had. And I've seen such a great like results while applying, um, some of those techniques into increasing the efficiency, uh, of our analysts and our ability to do things much faster, in much higher quality. Um, and that inspires me to just, you know, go ahead and see how we can broadly implement that within other departments.

Again, obviously being at Rapid seven and being exposed to so many MDR customers and sell customers really helped me to, you know, like I said, better understand, see the challenges, see the needs, see the market, and understand what a great opportunity, um, it would be if we could bring all this like, greatness of data science and the most recent innovations of AI into the stock, uh, and what great problems we can solve. Um, and that's what brought me into it. Very cool.

Alright, let's, let's pivot if we can now with, with Conifers, um, you know, started this company. Look, AI is certainly, uh, on top of everyone's mind, especially this week with what went on with deep seek and stuff. But what, you know, this isn't, this is ai, this is you.

When I look at AI and security, to me, there's two different pieces of it. One is securing against ai, right? The bad guys are going to use AI too, and we need to, to make sure we, we, you know, have a handle on that.

The second though, is leveraging AI to be more efficient in the security that we do. Conifers seems to me is more of, um, using leveraging AI to be more efficient, to offer better security, right? Absolutely.

Um, tell us a little bit about that. Yeah, for sure. So I think first we want to take, I wanna take a step back and just said, AI is great and everybody's talking about ai, but what we're really after is solving hardcore security problems.

So we've focusing on the business impact and the business outcomes on the things that we are doing, and we leverage great technology of making it happen. Like I've hear the buzz around AI kind of, you know, every company has the word AI to every product line that they currently have, but there is a little bit of dissatisfaction in the market with the outcomes that you see, uh, from some of those like products that you see out there. So we are, our main focus is the solve hard security problems through with AI and with data science.

Um, and what we are doing with confer is we are helping organization to achieve what we call solve excellence and be more efficient. And efficiency has been talked about a lot over the last, you know, 10 years in security operations. But unfortunately with all the different revolutions of the tools that you have seen, that we were able, we were never able to actually see the ROI and get to the point that finally we are able to unlock the efficiency gates, but also more effective.

What AI can bring to the table with the right combination with the individuals that you currently have in the organizations is the ability to scale up your coverage, making sure that you catch a lot more early warning signs, and you actually increase your quality and accuracy levels and scaling up in a way that you've never seen before. So what we're doing there with Conifer, it's we, we taking like holistic look at your stop as it's run today, we help you to recognize what are the best areas of opportunity that you have to be more efficient and effective, and start with a very, very moderated, um, engagement model that help you to have business success and save value. Um, bring those use cases and onboard them to your specific organization with a great combination between the models that we bring in pre-trained and your institutional and organizational knowledge.

So we take all of your incidents that you currently have in your SOC today, and we able to troubleshoot them and investigate them in more depth, in greater scalability, uh, which leads into efficiency, but also greater EFF effectiveness as a result of it as well. Excellent. Very interesting.

Um, you guys recently announced $25 million in funding. Yes. Yep.

Tell us a little about that. Yeah, so, you know, that has been really, really exciting. Um, you know, that's, that's came up as big trust from our investors.

Um, and we're very proud to have, like we take, like, take this amount of money and make, put it into good use. Uh, there are multiple things that we're going to do with this $25 million. And again, first thing is just our go to market in making sure that we, um, making ourself available to some of the great organizations that exist out there.

If it's service providers, if it's, um, enterprise organizations, we were able to go and access them. But on the technology side, we run a lot of interesting things. We have the cognitive stock, this, our proprietary patent pending platform that we created that utilized iGen ai as well as many out of flavors of data science to bring great results when it's come to accuracy, but also cost efficiency, um, into the enterprise.

Uh, so keep working on our platform, um, and also kind of accelerating our path with training more models and make them more efficient, uh, and use the, you know, most recent technology in order to bring that, um, bring that great opportunity to be more effective and efficient to a lot other different organizations out there. Agreed. Um, you know, I, I wanted this, excuse me, my tongue gets tied there.

Specifically, I wanted to mention the cognitive stock platform. This is a trademark, kind of, it's kind of the heart of what you guys are offering right now, and it, and it's really, you know, it's native AI and it's offered to organizations to solve the critical stock challenges at scale, right? Because, you know, I, I've been in the security world myself 25 plus years, and one of the companies I had helped co-found at one point, we really moved into an MSSP type of model, right.

We were, or, and I, you know, it's one thing to have a stock for a smaller company, a mid-size company, when you start getting a large enterprise or an MSP kind of model where you're managing multiple organizations, you know, every little problem is magnified at scale. Absolutely. Little, little problems become big problems at scale.

Talk about cognitive sock and how it helps there. Yeah, so there are few things that we're doing, and I think you hit the nail on the head there. Um, doing ai, it's, you know, great and do AI and small scale is great.

The ability to have consistent results in scale. This is something which is extremely important and consistent results in scale, which are unique to specific organizations. 'cause each organization is different with it.

Risk tolerance, with its assets, with its institutional knowledge. So being able to take that and make sure that the model are being fine tuned to those specific organizations, um, that's really key. And make sure that you can do it in scale across different organizations.

When you talk with a service provider or even, you know, a, uh, kind of enterprise organizations today, there are many service providers themselves with the different subsidiaries in the acquisitions that they do. So this is something which we invested a lot in, be in, in the process of being able to adjust ourselves to the specific technology and institutional knowledge that each one of the tenants or the organizations are using. We have this continuous learning that we do.

And this is another opportunity to say that, you know, humans are not going anywhere. And in order for us to, you know, we discovered a lot of information about a customer from this data, from the interactions with, with the analyst. Uh, but everything that we ingest into the baseline of each one of our tenants and customers is being audited and validated by human.

And why we did that, um, if you would be, if you would lack sometimes AI just to learn things by itself in certain areas without human supervision, especially when we know in some areas that you might have some bad behavior, you might scale bad behavior. And if you start ingesting small bites of bad behaviors into your models, you will scale up this bad behavior. And that, as you said, a small problem becomes really big overnight.

And that's probably would be the end of the AI implementation for that specific organization or service provider. So it's still important to have human in the loop. It's, it's still important to have human oversight and it's still important to have certain procedures and processes as doing just more information, especially critical information into the baseline of the models, uh, which are relevant to specific customers.

Being able to, whatever needs to be controlled, be controlled and audited. 'cause when things goes wrong, first, prevent things from going wrong. And if things go wrong, be able to nail it down, figure out where it is, and eliminate, um, what's going wrong as soon as you can.

So, super important. I, I agreed. Agreed.

Honor fruit we're almost out of time, but I wanna make sure we get this stuff in. Tom. ai.

Absolutely. And is there, like, what's the OnRamp, is there a free trial? Is there Yeah, you know, how, how do people get onto it?

Yeah, so we have, first, we, we do only, well, we sell our software only to customers that see the value in it. So we offer proof of concept and we would love to engage with organization and service provider that wanna come and see the value. Um, it's pretty easy integrating with the existing system platform portals and procedures and processes that you have in house, it doesn't take, uh, long.

And then within 30 days, um, usually, you know, most of our customers able to see significant value. Um, and our unique implementation model allows allow the customer to control how fast they wanna run and how, you know, how what is their readiness to go and adopt more and more AI in their organization. And as we develop that process of being able to, you know, let go for ai, it's also important that we stay on control of that process and making sure that every organization feel comfortable with the speed it takes them to enable ai.

Alright, Bob, thanks for being here on Tech Drug TV today with us. I appreciate you explaining all this to the audience. Best of luck with Conifers, come back and keep us posted as this continues to evolve and, and, uh, grow.

Thank you very much, Alan, appreciate that. All right, Thank you. Tom Findlay, CEO Conifers AI here, Aren Techstrong tv.

We'll take a break. We'll be back in a moment. Hello and welcome to digital CXO.

I'm Amanda Ani, and with me today I have Vene Pbu. He is the director of product at grapht. How are you doing?

Good, Amanda. Thank you for having me. Happy to have you on the show.

So today's topic is data assurance, but before we get into the topic, can you share a little bit about grapht and what services do you help provide? The grapht, uh, network as a service platform. And what I mean by that is we provide any, to any connectivity, whether it's from your private space to the private domain, private space, the cloud or, or, or the public domain and the services that you can consume.

So it's a consumption based model, so, uh, you don't have to worry about the network itself that's built out for you. So we provide our graph and backbone, uh, as a service. You consume bandwidth on that, uh, backbone to get you from point A to point B.

And we get you to the cloud. We provide you B2B connectivity. We provide, uh, provide you the data assurance service, uh, all on top of it.

And we, uh, provide you the SC van capabilities that, uh, uh, are very prevalent in the industry. So we have a bunch of services that we provide on top of this backbone. Alright, wonderful.

Well, so let's talk about data. And I wanna share, first of all, um, your stance is that data in motion has become one of the biggest, uh, security risk in the enterprise today. Can you explain a little bit more about that and why you think that?

So, data lives in three states. Like, uh, you mentioned data in motion. The data is addressed, the data is in processing, and then the data is most vulnerable when it's set in motion, because for the first time, it moves out of your private domain into the public domain.

So data address is sitting still in your domain, in your, within your parameter data. And processing is still within your parameter. It's when it exits your parameter and into the ether of the internet, that's when it's most vulnerable.

And hence, we think that, that, that's the most critical aspect from a security point of view of where data needs the most protection. With the industry going into the AI world today, the data is getting a lot more disaggregated. It's no longer just your data.

You need to share this data with your B2B partners, with LLMs, with GPUs, so on and so forth. So, uh, you are almost extending your parameter beyond your boundary with, uh, places where you're exchanging this data and graphene a data exchange platform. So with that in mind, we need to really protect this disaggregated data and extend your security posture and your risk appetite beyond just the parameter, uh, into the business domain of things as well.

So that, that, that's where aita sharing story, uh, begins. Okay. So can you share some of the top vulner vulnerabilities and risks associated with data and motion that business leaders should be concerned about?

So, data and motion. The, there are, uh, a bunch of things that they need to worry about, right? One, uh, is it encrypted?

Is it, uh, there's a spatial component and, uh, there's a temporal component to every conversation. So contextualizing data is first, uh, the first pillar that's key to protecting your data. So knowing what the intent of the data is, and then really plotting it on a map, right, of a, of space and time.

Am I consuming the data where it's supposed to be consumed? So let's say, Hey, I need to, uh, my servers, uh, are located globally, but if I'm authorized to only access servers in North America within sovereign boundaries, then that's the spatial component of where I should be consuming that data from and not going beyond those sovereign boundaries. There's a time component of it.

If I'm allowed to au uh, access data, I need to access it during work hours. If I'm accessing it out of work hours, that might be anos. So you need to really contextualize when you're accessing it, where you're accessing it and how you're accessing it.

So those are the three pieces of the puzzle that A, any CIO cso, uh, of an enterprise or a service provider would be looking out for in terms of how to protect and what to protect, uh, your data and motion for. Are there any tools or technologies that you recommend that, um, would help safeguard this data in motion Graph? I, I would recommend graphite and the, the data assurance offering itself.

Uh, what we've try, try to do for the first time is get security to govern routing, extending your security parameter into the network space. Uh, so what I mean by that is data is your most sovereign asset and giving that, uh, the treatment like any sovereign asset, uh, deserves, uh, data should stay within data embassies that extend your sovereign boundaries from point A to point B. So let's say you, you have GDPR data that's critical to you.

Uh, GDPR states that this data needs to stay within boundaries or within trusted entities that extend those boundaries. So keeping that kind of risk posture within your network of where this data travels from Point A to point B, who's the producer, who's the consumer, and those are entities that are GDPR compliant is key. Uh, and that's the control, uh, graph aim to give you, uh, with the click of a button, uh, of where your data moves, how it moves, and who produces it and who consumes it.

So would you think that corporate boards play a role in shaping policies for data protections? And, um, if so, how? Uh, both, uh, uh, corporations, uh, influence, uh, compliance, right, uh, needs as well as governments, right?

And these constantly keep adapting and changing. So, uh, having your enterprise ready and your network ready to adapt to these flex, uh, or these modifications on, uh, data governance and compliance is key, right? It's hard to, uh, rip and replace infrastructure, say, uh, a government changes in a region, the compliance and regulation or encryption regulations change.

It's hard to rip cables apart and reroute it around that region if it's not compliant to your, your enterprise. So having a programmable network that can really just add a click of a button reroute that, uh, path around that region is what's key. Uh, so it, it's a mix of both.

Uh, the regulations are the compliance needs of an enterprise. Example, financial in institutions have PCI kind of regulations. Governments may have the GDPR and HIPAA kind of regulation.

So, uh, but, uh, businesses have to comply to both. So the, it's a mix of both and giving you that adaptability is key and giving you that control back is key. Wonderful.

Well, um, out of all this, what is, um, what is say one key takeaway that you could leave our audience with today? The key take takeaway is, uh, just protecting your perimeter isn't enough. Having a network that extends your segmentation from your boundaries, from your edge, uh, really getting your business, uh, traffic, the business internet.

Uh, so the internet was inherently built for, uh, open communication, but enterprises need a different kind of internet. That's the business internet and graphene aims to provide you that with its, uh, uh, backbone as a service offering, allowing you to program the business internet based on how you are consuming and producing, uh, data and how you govern that data is key. So I, I, I would really encourage folks to focus on how security can really govern the routing, uh, in, in the age of the business internet.

All right. Well, thank you so much for coming on and sharing your insights today. Thanks.

Thanks for having me, Amanda. All right. And thank you to our audience.

Stay tuned. There's more. Hi everybody.

Welcome. We're glad that you've joined us today for another episode of the latest greatest cloud transformation late great cloud transformation. We're talking about really sort of the next generation of how we think about the cloud and the things that we're doing with it.

We're talking about security today, about safeguarding innovation and, uh, strengthening that security will be jumping into app, app security, and a lot, a lot of things here. But, um, before we get too far down the road, thank you for joining us for this video series. Uh, the, the last great cloud transformation is sponsored by CloudFlare.

We're glad to have them, uh, on board with it, with us working on this, uh, helping input with some topics and things like that. And obviously participating on, on our, uh, live editions, which we do on a monthly basis, as well as these recorded episodes. So, thank you for being here with us.

My name is Mitch Ashley, I'm VP and practice lead with futurum Group, analyst firm, uh, heading up the analyst area for DevOps, DevSecOps, application development, AppSec, et cetera. So kind of right in, in vain with this, uh, my co-host Alan Shimel is, uh, unat, uh, de detained, or whatever the word is the phrase is. And, uh, so I'll be, I'm, I'm hosting both parts of the chair today.

Uh, you know, it's a little bit of a coup, but he'll be back next time. We'll see him on our next episode, I'm sure. So let's get to our conversation, to our topic.

Um, let's first start by doing some introductions. I know Chris has been with us on a few episodes here on some different topics. You've been on other webinars with me and talking a lot about application security and, and, uh, cloud Chris Blas, introduce yourself.

Oh, I've been a for company my way through the security industry for 30 something years. Uh, I inflicted an early firewall in the markets, something called border wear, uh, in the early nineties, and ran Cisco's firewall business, the turn of the century. I've been following this inevitability curve and my new series on Huron Textron, um, from one spot to another, from firewalls into, uh, sim and network management.

From that, you know, the obvious next step is threat intelligence. So I, uh, chaired an ISAC for a while, and, uh, supply chain has been my focus the last five or six years, you know, so, you know, software, bill of materials, hardware, bill of materials. How do we connect all these things, which a, a and, and currently, so, currently I'm, my main role is I'm vice president of strategy for sbe, which is involved in the SBO space.

And I've been, uh, co-sharing several, uh, cisa uh, working groups on SBO m sharing. So we're currently have a group looking at ISACs, um, as SBO distributors. How does that know in the middle start taking this information and, and propagating it SBO software bill of materials?

Absolutely. Great. Thank you, Chris.

Um, Katherine, Katherine, welcome. Glad to have you on, I think the first time we've had you on the show. Katherine Newcomb with CloudFlare, please introduce yourself.

Yeah, great to be here. I'm excited to talk about application security. Um, my name's Katherine Newcomb.

I live in Denver right now. Um, I've been in cybersecurity for about five years at this point. Um, and I started in the network firewall space, um, an encryption, and now I'm a product marketing manager for CloudFlare, um, for their application security business, uh, where I focus on their web application firewall product, um, our software supply chain product, as well as our encryption and certificate lifecycle management products.

Very nice. And, and I do like to say full disclosure, Textron is a customer of cloud flares. We do use their services.

Enjoyed very much. So thank you Catherine, and team for that. Uh, last but not least, another newcomer to our show, Kurt Handel, who's with, uh, Teradata.

Tell us about yourself, Kurt. So I've been working in security probably eight or nine years at this point. Uh, but in the software industry for close to 15 years now, anywhere from development, uh, into business analysis, product management, even, uh, doing a little bit of red teaming myself.

But, uh, I am currently the chief security architect at Teradata. And so I've been focused on architecture mostly for the past six, seven, possibly eight years, and really kind of a generalist. So AppSec is where I spend the least amount of my time, but we focus on the architecture, the requirements, threat modeling, um, especially compliance.

We do a lot of the, the major compliance frameworks at Teradata. So we've been pushing that recently. Um, and I'm based in the Pacific Northwest, up in the Seattle area, and happy to be here.

Very nice. All the weather and fires and it's cold, and I'm just glad we all made it. Maybe it's 'cause we didn't have to travel anywhere, so, so I hang tight.

I'm glad we're all here. And you know, our, our thoughts go, our hearts go out to the folks dealing with the fires and, and, uh, some weather down south and southeast, et cetera. So, um, let, let's kind of jump in this way.

Um, it, it's a big topic when we talk about sort of the kind of current state of the cloud and where it's moving to. Um, but I don't think it's too much news to everyone that application and app APIs, API first kind of design into applications, you know, it isn't just things that sit at the edge anymore. We think about also the security of the apps and the kind of, uh, software we're creating, the innovation that we're making, um, as maybe as part of the cloud.

'cause sometimes application lives within it, you know, like a, like a provider like CloudFlare or certainly at the edge or at the core as well. Maybe Catherine, if you wanna start us out with, how do you, you're, you're, you're managing, doing product management in this space. How do you look at this, uh, sort of this problem or this space and define it?

Um, so looking at application security, um, when we're talking about this at cloud, there, we're mostly talking about web application and API security. So if you're an OSI person, layer seven model, um, and you know, when people are accessing these external facing web applications, they're doing it from a ton of different devices and in a ton of different ways. So they're accessing from things like mobile, uh, desktop, laptop, and they're accessing these apps that could be hosted anywhere.

So on-prem, in public clouds, private clouds, hybrids. Um, so as we're securing, we need to think about how can we secure, um, all of these users and the end servers as they're sort of accessing these web apps, right? So how do we make sure that, um, mobile traffic is protected, user data is protected, um, and sensitive data is not, you know, leaving an app.

And then how do we make sure that a web app server itself is protected? Um, so at a very high level, that's about what I think, that's what I think about when it comes to application security. Um, some new things we're thinking about in this space.

Um, I talked about software supply chain. This is increasingly becoming, um, an area of interest as people create more complex apps with more third parties in them. Of course, API first development has also meant we've had to adjust our thinking a little bit around application security as well.

Kurt, how about you as a, as an architect, security architect. May, maybe you don't get into the innards of applications per se, application security, but traffic over there. Obviously our networks are heavily API driven.

Um, you know, when you think about the security architecture, where does this fit into your purview? I think it, it fits in really everywhere, right? So we're, we're building these huge applications, sometimes small applications.

I mean, we do all sorts of scale at Teradata. And in my previous roles, I've, I've worked with pretty simple apps all the way to super complex microservices architectures. And so, like Catherine kind of said, you have the mobile aspect, you have the server, there's application code literally everywhere, including on the person's device.

And so how do you secure it as best you can, um, within reason, right? Because if it's too secure, it doesn't work. If it's not secure enough, well, you end up in the Wall Street Journal and you're in trouble.

Um, so we, from an architecture standpoint, we really try to focus on all different aspects of it, where the biggest threats lie, um, and then implement controls and use technologies to, to simplify the implementation and streamline it without making it overly complex. And so it's, it's just becoming more difficult given that, um, the, the kind of classic perimeter is gone, right? I'm sure you can relate to that, Chris.

Oh, yeah. Well, it was easy back in the day, right now, you had to get on the internet and you needed a firewall. Get a firewall, right?

And I'm thinking as Kurt and Catherine, your comments remind me of these transitions we go through. Like there was the mainframes before our time, but you know, I, I'm old enough to have seen the end of that where all of your capabilities are just to keep one computer running and run terminals and printers and things off that. And then we get into, you know, or where I came in, where we're starting to build networks, fractally more complicated.

Just, you know, how do we do that with, when all of our resources, were just keeping one computer running, we figured it out, you know, now we're here, we're talking about web APIs, Catherine, you know, you know, the data going in and out on with being stored 30 years ago, you couldn't have that conversation. Now we're saying, alright, what do we do in this case? And it's very complicated.

And I think in, and Catherine mentioned the supply chain, this is, I think we're filling in the dots. Security has been, is not, i i is not new, right? People have been saying, you should know your inventory for a long, long time, and we've gotten away with not knowing it.

Now we're starting to fill it in, need to actually know where the software is, where the data is, and we're working through that. So it's exciting times, but it's not different in type than other transitional periods. Certainly is an evolution, right?

Of what we've gone through. And to think about, you know, from the bas and host days, early, early on, free firewall, um, well, Firewalls used to be a million dollars a year. I think about I got involved, you know, at least as I tell the story, there were a hundred in the world and they typically were seven computers and a team of people.

And my argument at the time was my mom needs one. Yeah. You know, and so we're at this stage where what used to take so much time in here in the API, uh, world has to take less time a lot.

It, it, so let me, let me throw out this hypothesis here. I think it may be pretty obvious, but maybe it isn't, is I think we live in a world, you know, now we we're thinking about things as zero trust, right? Of, of you, you know, anything is susceptible, being compromised and could compromise other things.

How do you pre protect all parts of the network applications, the infrastructure? But we're also living in a world where if so much is determined by what our applications do, not just connecting users to apps, but applications really utilizing the network, being part of the network. It's a dynamic world, right?

It, it isn't a good set of firewall rules and an application firewall, and we're all good, kind of set that up. And it isn't the old days of I've got a pizza box in, in my rack for every function that I need, and they're all doing their thing. I'm good, right?

We need it. It's a much more dynamic environment. So I'm not saying we're reconfiguring our security all the time, but a security has to adapt to, you know, what's happening in the application.

Because we may distribute it to a different part of the edge tomorrow with Kubernetes, or we may, you know, uh, acquire business and suddenly our network has looked much different than it did, you know, three weeks ago. I'm, I'm curious, Kurt, as a practitioner, you know, how do you think about that of, you know, you mentioned microservices and all the things that are being created, you know, in the groups that you're working with. Um, we, we hate for security to be sort of the last thing to be thought of, but you wanna be in the conversation so you can prepare as well as react when you need to react.

I think what you just said is, is really important. You wanna be in the conversation. You don't wanna be doing this retroactively.

And so when you're, when you try to tackle security retroactively, it is infinitely harder to accomplish than if you do it from the beginning. So I have, I do it both ways. I have teams that we work with proactively where they bring us in at the very start and we're building the design with them shoulder to shoulder, drawing the picture in doing security by design or by default as we like to say now.

Or we have legacy applications, which you're doing retroactively, and they're quite a bit higher in terms of risk because they've been neglected for so long. Or we find out about something after the fact and it's like, well, how did this get out there? Well, there's shadow IP in a lot of the world.

And so it's, it's hard to, to really kind of put a, a recipe together that successfully achieves it. And then with the, the rapid pace of technology today and how the cloud has just kind of blown this wide open where people can deploy new applications in a hundred different ways faster than ever. How do you keep up?

So you have to implement tooling within reason without doing, without having too much sprawl. You have to have the right personnel partnering with these teams, uh, to ensure that you have coverage and that you, you're really architecting things from the start. Um, and not just kind of using band-aids and bubble gum per se, to, to secure your environment later on.

Catherine, appreciate your thoughts on this because, you know, I remember the days of networks for speeds and feeds and points of presence and connecting A to B and kinda looked like this nice diagram that you stitched together and that was a network and you secured it. Now it's overlay on top of overlay and it's changing and, you know, it's, it's multiple pieces that, uh, much more complex to, to secure. How do you, how do you have this conversation with people?

Yeah, definitely. So as you were sort of talking about this, you know, obviously there's a need for responsiveness and customizability and security, but I actually also wanna make the argument for unified policy management in application security. This is something that I've seen actually, for example, um, we have some customers who have protected their SaaS apps, like what is traditionally more of a network firewall or zero trust type use case with the same policy they're using for their web applications.

And by doing this, they're able to do things like make sure that zero day exploits aren't able to exploit their SaaS apps, you know, as well as their, um, web apps. And we see a lot of value out of these unified policy managements. I was talking earlier about, you know, how we have all these apps hosted in different places.

We see a lot of customers, for example, will host, um, you know, an app across multiple clouds for like a resiliency use case. If they're worried about outages, you'll, you'll certainly see that, um, for example. But then how do you have to, you know, actually secure an app that's stored in multiple places?

Do you write different policies for, for wherever those are stored? Um, do you write different policies for APIs versus, you know, traditional apps? Um, so we see a lot of benefit out of like a unified policy for all of those disparate sort of endpoints and all of those disparate, um, locations that they're stored.

Uh, for CloudFlare in particular, how this sort of works out is our WAF is like the backbone, the architectural backbone of the rest of our application, um, security portfolio. And this works out really well because you can do things like have a WAF and an API like positive security model protecting your APIs. Um, so you could do things like detect zero days and volumetric attacks, which are, you know, APIs can also be susceptible to as well as, you know, do the things like Ebola and, and all those API specific attacks all within sort of one, um, control plane, which we find a lot of people get a lot of value out of because of this really, really disparate environment.

Okay. Chris, I saw a lot of hand waving head nodding you about, jumped outta your chair on this one. And so I kind of have feeling you might resonate with this.

No, um, I, I gotta throw out there, I was gonna, uh, before Catherine got into the, the policy thing ask wearing has been on my time, but yeah, the concept of an sbo, the software builder material for the current release version of Adobe Acrobat as opposed to an SBO M four as we're look talking about here, some ephemeral web app that one time for five seconds exists in the cloud. You know, think about that. How do we kind of deal with that?

And I, and, but I think policy is, is the answer all hacking? All hacking is policy hacking. I will figure out how you do things and I will figure out where the gaps are and I'll engineer that gap.

And we live in a world right now where we generally have no idea what policy applies to any of us anywhere, with few exceptions. And in this topic, and because I'm used to the supply chain topic, imagine I needed to get the, the SBO M or custody information about a piece of software on his phone right now. I could get it in between five days and six months today I need to get it in half a second.

That means I need to read the policies between me, the person who bought the phone and the first time the company I bought it from, and like their relationship, their contracts, their policies, you know, upstream all the way. And we have to get that done in the next decade. So without unified and, and, and adaptable, you know, transparent policy frameworks, none of this technology is gonna make a difference.

So I think we, we will do that. And there's interesting things going on down that path. It's kinda interesting in a way, just connecting dots between what you said, Catherine and you were talking about Chris, there's your own unified policy management, right, of what you're doing.

So you know, you're, you, what you're applying where and how you're applying it, and then that's how that interconnects or interrelates with the people you connect with, work with, use their service product, whatever that is too. And I, and I appreciate what you said Chris, about, think about just serverless technology like a lamb to kind of service, right? That, you know, it's there now, it's gone tomorrow may not be the same thing it was a second ago when it, when it ran.

Um, so, uh, in, in some ways, Catherine, it's all sort of a dynamic unified policy management, right? It can't be a static thing. Am I, am I on base here?

Yes, of course. You know, you do have to be responsive to the environment, um, you know, a threat landscape. Um, this is one, one area where I strongly advocate for actually ML driven, um, detections and policy.

Uh, this is a thing where, for example, if you have a really large data set, uh, you can train your ML models. Um, how we do this at CloudFlare, just 'cause I think it's a little easier if I give an example and it's, uh, we will score each request on a scale of like one to 99. And if something is less than 30, that means like it is very likely to be an attack.

And because we have, um, hundreds of terabytes of requests, or sorry, hundreds of millions of requests every single day, um, we have so much data we could train this on and say a little blog in Malaysia gets attacked by a new attack we've never seen before. Suddenly because that tiny blog in Malaysia got attacked that gets feed in fed into our ML model. We don't have to rely on a security engineer to like go and find and analyze that attack and turn it into a regular expression like firewall rule.

Um, the ML will basically just say, okay, like since it matches something like this, um, we will just automatically block it. And this is why I'd say ml um, sort of combined with that traditional, um, you know, security analyst looks at the traffic and writes a rule that matches it and then blocks traffic. Um, you gotta combine I think these types of approaches.

So ML is a really, really great application, um, when it comes to being responsive to the threat landscape. And we have some data around this as well. Um, we recently, not that recently, like half a year ago released our annual application security trends report.

Um, and we found out that, uh, for example, like zero day vulnerabilities, um, we probably wouldn't have been able to find this out with just security engineers analyzing it. But with our ml, we were able to detect, um, and exploit 22 minutes after the, uh, proof of concept was posted online. So, um, really, really great applications there.

A lot of interesting stuff going on for sure. Well, if that doesn't make the case for dynamic security, what does, right. Um, I'm curious, Kurt, how do you, is, is someone, you know, applying these things, applying security?

Are you, are you looking at things like ml? Are you doing any via yourself? That's something you look for in the vendors, the partners that you work with.

How do you leveraging either that or other kind of technologies to help shorten that cycle between when things change and how you can account for it and secure it? Right. The, I think the ML piece of it is, is hugely important because I mean, humans, we're slow.

The, the technologies we use, the, the computers and I, each servers process all of this far faster than the human brain and I ever could. And so we need to augment ourselves with this technology. So anytime we're evaluating new solutions and bringing them in, like I'm currently in the process of implementing a big one right now that focuses on platformization and ai, ml, it's all part of it because it humans with eyes on glass, like it's great to have those guys in the sock, but they'll get overwhelmed very easily with the speed at which things happen today.

And so we need to leverage technology and machine learning enables us to do this faster than ever, and it's only getting better, right? And so augment the human with that technology and you can very quickly pare down all of that information to what matters most and focus on real attacks like Katherine was just talking about. I wonder, you know, there's so much activity around ai, of course, a lot of it because of gen generative ai, um, Chris to does security engineers have to become machine learning experts to be able to do this stuff?

What does it take to really leverage it? No, but knowing, knowing something isn't gonna, um, uh, causing any problems. But, uh, i I just couldn agree more with, with both, uh, with Kurt and Catherine.

'cause you know, and, and you're point Kurt's all about time, time to transparency. How, how long, and again, I've seen this over and over in my career where we get to these points where what we're mostly doing is sharing the war stories. You know, I have no idea it was 72 hours, none of us slept.

There was caffeine. And, and my my question always is, okay, if there was twice as much, what would you do? Because obviously that we're at the limit, we can't possibly work any harder or stay awake any longer.

And, and this, yeah, ai, ml, Oracles, whatever we call it, this, uh, my, a big has been a big part of my, uh, my focus on supply chain before it would, you know, AI became, you know, uh, a general, um, uh, generative, what the hell do we call it? I'm sorry, I forgot. Yeah.

Generative Ai. Yep. Generative ai.

Yes. Too many terms To throw Around. Yeah, because again, we need to, you know, just for supply chain things, I need to read the contracts.

I mean, I can literally call someone up, you know, it's not a security engineer, but it's some administrative person of the company and I had to get them on the phone and get them to pull A-A-P-D-F and read the contract and find out if the clause allows me to get the information I need. That's not worth a human's time. I mean, that's the kind of stuff that computers can do really well, and they're just beginning, but that's obviously the direction we're going.

And if you can't see your policy environment five years from now, by various definitions, your competitors will be so much faster than you are that it won't matter anymore. Cur, I'm, I'm curious, without giving us too many specifics about Teradata, I'm not asking you for that, but what's your sense of, what are the, what are the new priorities that are on your Yeah, on your horizon or things you're dealing with now and that you've kind of added in the last year or so? What's changed about how you're thinking about security and that you've gotta address now?

I think there's, there's always classic problems that we, we have to deal with and tackle. Like, we can't forget things like identity and network security and the rest of it. But the, the prevalence in the emergence of generative AI and putting AI and machine learning in everyone's hands has meant that security teams have to be hyper aware more so than ever because these new technologies, people are latching onto them without considering the risks.

They're like, that's awesome. I can speed up everything I'm doing. And suddenly you see a new story about, well, what was it like Samsung engineers leak their code through a generative AI solution or whatever.

So you're, you can quickly lose intellectual property or put it at risk. And so we have to think about securing our environment for those solutions, or putting the guidance out for people to use AI and machine learning. Um, and I mean, getting visibility of all of this, and another big one that's been getting pretty popular and we're seeing a lot from different vendors and acquisitions and whatever, is data security, posture management.

Where is my data? Where is it moving? How secure is it?

Because at the end of the day, that's what the attackers want. They don't wanna sit in your network and use your resources to, to launch attacks as much as they used to. They wanna grab your data, steal it, monetize it.

So need, we're, we're focusing on data security big time in, in the more recent years, especially, um, forward looking because we have more data than ever. Interesting. Catherine, from your perspective, you know, communicating with so many companies, what are some of the changing priorities from your, from your viewpoint?

Yeah, I mean certainly the gen ai, um, piece is something we're seeing a lot. Um, everybody wants to put an an LLM on their web application. Um, and of course that means that you have to think of that as like a data security concern as well.

Um, because you wanna make sure your LLM is not gonna like accidentally leak somebody else's social security number because that's certainly happened before. Um, and so at, at CloudFlare we're thinking about this of like, basically how could you basically just put a WAF in front of an LLM, um, from that perspective, how could you prevent it from exposing sensitive data to the end user? Um, but then, you know, you gotta think about these more complex issues as well.

Like how do you prevent somebody from poisoning the model? How do you prevent, um, you know, some of these other, like how do you prevent it from hallucinating? Uh, these are all, you know, sort of adjacent to security concerns.

But, um, but nonetheless, we see some security teams focusing on this, um, increasingly. Um, additionally we also think about, you know, the, the LLM sort of security use case as a little bit of a just, um, increased API security use case since a lot of times, um, people are not building these LLMs themself and hosting them themselves. They're often, you know, bringing in LLMs from third parties, which, uh, necessitates, um, APIs, right, for integrations.

So how can you make sure that these APIs are staying secure and not leaking them back to the host and whatnot. Um, so that's definitely something we're seeing as well. Um, I would say additionally, one thing I've been hearing a lot lately is, uh, software supply chain security.

Um, I think Kurt mentioned the beginning, um, sort of securing code that lives on the client device as well. Um, this is something that we've been hearing a lot about, especially as it comes with the PCI four, um, compliance, which is gonna be mandated at the end of March, um, in a couple months. Um, PCI four has a new compliance requirement around client side security and securing, um, the client side, like software supply chain.

Um, so this is something we've been getting a lot of questions and inquiries lately. Um, you know, how much are organizations responsible for, um, the code that loads on their end users devices, uh, when they visit their websites? Um, this is something we're seeing a lot of people trying to actually actively get control over, um, and make sure that they're not, you know, serving, uh, code to the client devices that could do things like download a crypto mining software onto their phone, which, um, believe it or not, we have seen somebody's trying to make, you know, personal laptops part of a crypto mining network, which is pretty crazy.

But, um, so yeah, I would say the client side component is, is something I've been hearing a lot lately as well. I, I just have to say, I, I love living in a world where we can use the term, uh, you know, hallucinating artificial intelligence in a conversation like this. Seriously, just, we, we understand about that.

It's not a sci-fi movie. It's real. Oh, it's, it's real.

Yeah. Hey, so I've, I've kind of a left field question for you, Chris. So if this, if I throw you too far off the track, I'm guessing you're thinking about this though, is, is there an SBO m in our future for LLMs and s SLMs and all of these things?

'cause in a way, this is a whole nother part of the software supply chain, right? We're handing off to something that's doing inferencing, either on a chip on our handset or in the cloud, all of the above. How does that fit into, do we need to be thinking or at least wondering how we're gonna solve this problem And not only in not the left field, and that's, that's right in the middle of the, the track.

So in short, yes. You know, there's ano there's another working group, um, Dmitri Rayman, uh, my colleague CTO at at SBE is, uh, a co-chairing now on, on AI bomb, right? An AI bomb has been talked about for a long time.

So what does that even mean? You know, so AI is code, so there's this, you know, same sort of standard SBO stuff about that, but it was also the training data and the models that produced. Right.

And this sort of goes back to my last comment about ephemeral ephemeral SBOs. You know, we start with the idea that I am a software provider and every 16 years I release new code and I carve a new SBO m you know, on purist graphite. Um, but we live in a world where code gets compiled and used all over the place.

You know, how do we even look forward and say that I can commit to a policy that says I will, if asked, provide the contents of this code without, um, actually going out and printing or saving or producing quadrillions of SBOs forever, you know, in, in exabyte storage. Uh, so this AI is, you know, what we're currently calling AI is just another forcing function of the level of complexity we're at. So we need to be able to provide the answers to live up to the policies that we've agreed to, um, which is, you know, you know, in the SOM case we're talking about a software inventory that I will be able to tell you what code that was running or you know, what data set was used, and we have to get there.

And, and, and it's, it is reasonable progress down that path. It's a, it's a complicated one that is very similar patterns to how we'll do other things of similar complexity. Um, Kurt is, is that on your radar yet at all, kind of thinking about security of, from a supply chain for LLMs and AI and ML algorithms and all that kind of stuff?

No, I mean, it's, it's certainly jumped up on the radar, especially since the whole SolarWinds thing happened. Um, as Chris was talking, it got the wheels turning in mind of, well, if we're gonna be kind of, we're moving towards leveraging a AI in the sense and dynamically generating SBOs and things, is this another attack vector we potentially have to watch out for? Is how do you weaponize that and, and protect against it?

Because I mean, as we see attackers evolve their tactics and techniques faster than ever, they're coming up with new creative ways that defeat the traditional approach in microseconds. And so how, how do you stay ahead of that curve now in, so I obviously, I don't have the answer right now, but it's, it's really interesting as Chris talked to start thinking about this, this new sort of problem that we're facing. And again, it all falls back to the rapid evolution of technology.

Yeah. Speaking of that evolution, uh, just in the last week or so, uh, Satya Nadal, the head of the Microsoft was talking about the death of SaaS, meaning that's kinda the clickbait one liner. The, what I think he was really talking about is e evolving nature of software architecture that I would describe it as.

Today's m services or backend code are tomorrow's AI agents, right? We'll see more and more parts of apps built through, you know, with or through or maybe completely with AI agents. And it reminds me of going into the, uh, cloud native era of, oh, how do we secure microservices now that we're gonna do that kind of thing.

That's kind of the, that's the next edge that we're, we have to work on and think about how, uh, there are different things we have to do for securing AI agents. How are they orchestrated? Is it Kubernetes or it, some other thing that's managing all those things.

And, uh, given that we're putting AI agent building capabilities and everybody's hands, in many cases, it, uh, could make for interesting. I use that in a nice way, uh, interesting environment to try to secure and manage. So in some ways, the future is bright, but it may be a pretty intense, at the same time, same time.

Well, I think kind of building on that too is the, the technology behind ai, it's backed by machine learning. Like you're, you're making technology autonomous, right? So it's not as predictable anymore.

So how do you secure what, when you don't exactly know what turn it's gonna take next, Non-deterministic. Right. Well, I, I gotta add a note, a note of hope though, because it's easy, you know, to your point, uh, Kurt, the short answer is yes, because there's a new attack vector.

Oh, yeah. Um, but, you know, you know, throughout my career, I've been already this one, it's like, we'll probably keep the lights on. It's like, no, no, if we don't do this and that, then you, we will, you know, but the, the, we're on this, we're doing this call right now.

We've managed to figure out everything else over this point. And not only that, but I think that we're, we've been mowing the lawn. I think, you know, what we need to do, generally speaking in cybersecurity has been known maybe forever, certainly 50 years, but we haven't gone around to doing the vast majority of it yet.

'cause we haven't had to. But as we do, and I, I will take a risk and, and put a lot of my, my faith in policy, you know, in, in real policy transparency, you know, in, again, in this decade it gets harder to be an adversary because, you know, these are the happy World War II fans out there, you know, or no fans, you know, but the, the ubo wars, right? There was the happy days when you could just have a U-boat and sync shipping all day long.

You know, that's kind of, most of the world, most of the, the history of the internet to date is not necessarily gonna stay that way, that long forever, where there's always a new attack service, and there's always a, a, a new way when the last one is, is locked. I think we will, we'll keep it running. We will all be fine.

And I think over, you know, at least over a period of decades, being an attacker will become much, much more difficult. I mean, I might argue it already is becoming more difficult. It 'cause the, while, while the, the technologies we use as practitioners are getting more advanced, that helps make it more difficult for the adversaries of the world.

But that's not to say that they can't employ similar technologies, right? So now we're kind of, we're creating that chicken and egg problem all over again and playing a game of cat and mouth. It's kind of the next arms race, if you will, as technology evolves, everybody has access to it.

Well, let's do this. I appreciate all the conversation and we brought up a number of topics, um, just as a kind of concluding thought. Uh, we, we've been talking about what are the things we need to be thinking about?

Maybe they're new on, maybe they're on the horizon. Maybe you're already working on this today. Um, if you had to say, there's one thing you'd really want to emphasize this, if you were, you know, somebody who's listening to this and maybe making a few notes, the thing that sort of stands out to you as something really important to be thinking about in the next, let's say six to 12 months, if not today.

Um, Kurt, do you want to give us your thoughts and then Kathleen, if you would, and Chris, you can wrap it up for us. Sorry, did I say Kathleen? I mean Catherine, excuse me.

Kathleen. I work with a Kathleen. Sorry.

I've been doing that. All good. Okay.

Yeah, I, go ahead, Kern. I mean, it's, we wanna avoid that situation where everything is a priority, so nothing's a priority, right? I think we, throughout this conversation, we've highlighted the importance of ESMO is we've highlighted the importance of application security and how it's, it's becoming more important than ever because our application code is, is literally going everywhere.

And that's, that's kind of the gateway for a lot of the attacks we're seeing in the world today. And so I think the, the emphasis is on application security, but it's also to say, let's not forget the rest of it, because all of the, the other parts of cybersecurity are hugely important. And we still need that visibility.

We still need the coverage, and we need to be thinking about ease of use as well, and avoiding the sprawl. So I know these aren't necessarily specific cybersecurity things, but they, they help you simplify your approach and, and focus on what matters. And that depends, that, that changes everywhere you go.

Every enterprise or company has different priorities. And so I think focusing on those things help enable us to, enable us to focus on what matters for where we're at currently. You're good, Catherine.

Yeah. So I mean, like Kurt said, you know, we wanna make sure that we're not making everything equal priority. So I think when it comes to application security, which is of course, my area, what I would say is most important in this space is visibility.

Um, the attack surface is getting more complex, applications are getting more complex. Um, you know, where they're hosted is getting more complex. So how do we actually have visibility into our entire, entire application attack surface?

How do we have visibility into the APIs developers are creating so we can actually secure them? How do we have visibility into the software they're adding, um, to these apps? Uh, that I would say is probably the most important thing for application security and also one of the most challenging things.

Excellent. Chris. Uh, then, you know, Kurt and Catherine both want exactly where I'm going, so I'll just build on that.

You know, do do things that save you time to transparency. You know, if you, you know, don't panic, nothing's on fire. And, and when things are on fire, panic less, right?

Just take your time and, uh, getting visibility, you know? Yeah. Look at how long it takes you to figure out.

And anytime you find a, a, a way, you know, in this, in this topic we're talking about here, to spend less time to figure things out, you have all that time back to do things. And it's easy to just, you know, particularly in transitional periods, to just do more and more and more what you've been doing, you know? But, uh, understanding the environment you're in so you can apply your resources appropriately is, is everything.

And there are lots of ways to do that these days. You know, there's, there's a lot of Russian panic and there are a lot of, and you know, I'll say it, AI and things like that out there who will actually make your life easier, give you some of your time back. Mm-hmm.

And you'll point, feel better knowing what's going on, make, make, and make better plans and better strategy, Uh, to that point. Exactly. Chris, and, and Catherine mentioned it around, uh, ml, you, some of the things that I'm really excited about AI is actually just the understandability of what's happening.

You know, Kurt mentioned about as things ramped up or, or you did, uh, uh, in, in the, if the tax doubled, right, how would we handle that if we're already maxed out? So some of it is just handling the volume of things that are happening. But I think one of the things that I think is most exciting about generative AI is it's also so complex.

No one person can understand the full system, right? Or maybe even understand truly what's going on in a case of an attack or where you have vulnerabilities. And generative AI is starting to make some inroads and helping us understand systems and, and giving us some insights to some of the complexity.

We may not be able to fully get into our head all at once. So for example, I've been doing some work around how do you modernize mainframe applications? Well, nobody was around that built those things.

Well, maybe it's people that built the network aren't even around, right? So help us understand what really is happening with all this data that we've collected. And the natural language interface through that is, is a great aid.

And I think it's just a real practical thing that we can start to begin to use today. So don't think of AI as just as the next, you know, it's gonna replace all of our software and it's all gonna be different. And what do we do?

There's things today that it's already helping us with. So, you know, there's some real things too, not just what's on the horizon. Well, thanks to all of you.

It's been great, Catherine. Uh, we appreciate your perspective, and Kurt, you're bringing, um, your experience and perspective. And of course, Chris, always good to be chatting with you and your connections into the security world.

And some of the folks are working, collaborating together, which by the way, is another superpower we have in security. And that's the fact that we work together and collaborate on, on these things. We're not going at it alone.

So thank everybody for the good work that we're doing to help advance. We hope this has been a helpful conversation for you in thinking about the, the last great cloud transformation, what we're doing differently and thinking about, uh, as we move forward. So as we've got our heads down, getting stuff done, getting our priorities done, getting our plans in place, and executing for 2025, but also kind of thinking a little bit about what's next and what we might be considering and learning from others that are working in our space.

So, thanks to all of you. Thanks everybody for joining us today, and thank you to the Cloud four team for, uh, for sponsoring, um, our show today. And we look forward to joining us either on another recording or Sure.

And check the calendar for one of our live events where folks can ask questions and engage with us in a similar kind of conversation. We have many of those coming up. We'll talk to you again soon.

Take care everybody. This is Textron tv. Hey guys, thanks for the throw.

We're here with Dale Hulk, who's senior director of Information Security at Red Scale. And we're talking about a new continuous controls monitoring report put together by the CISO Society. And well, that really just goes to the whole heart of this, what's going on with GRC and cybersecurity.

Dale, welcome to Shell. Thank you, Michael. I appreciate the invite.

Looking forward to having a conversation. So give us the high points of the report and why you guys think that this is significant, because, um, there's just a lot of things to worry about in the land of cybersecurity and compliance. So how do we make this a priority?

Well, I, I mean, the, the key factors around the report that kind of surprised me was 95% of CISOs don't believe that they have a good control on, on their, um, on their programs. And, uh, but outta that 95%, they all believe that that automation is the key, right? So when you start talking about GRC, I don't know about you, but in my head, I hear antiquated processes.

I, I hear, uh, spreadsheets, I hear right? Siloed data, I hear people who can't have conversations. Your data is stored in, you know, disparate places all over the place.

What I hear con continuous controls monitoring, right? That is more of, you know, uh, dynamic operational control assessment, right? I have, I have a security tech stack.

I take the outputs of that security tech stack, and I get good compliance, right? And I do that by optimizing my resources, right? It's a force multiplier, really is how we look at it.

How do you keep people from wasting time, right? How do you keep companies from wasting money? I would say, you know, the report, it cited a good, um, 40%, uh, either didn't have funding for compliance or it wasn't a priority, right?

Where I, I, I'm sure if you talk to some of these, uh, some of these, uh, auditors, right? They understand that, that in order for them to operate anymore, right? Compliance is, is gonna be a priority.

And how do we achieve that compliance without crippling security? That's the real question. Everywhere I go, somebody is complaining about regulations, and yet it sounds like, you know, we're not doing anything to automate the process to make it less painful to execute.

So are we involved in some sort of catch 22 here? Absolutely. Right.

So there's, I think there's three pieces of that. So first, nobody can agree on a standard. You have Osco that's been around, but it hasn't been used because nobody can agree that that standard is worth it.

You have OCSF, right? Um, you, you know, the, the, that, that chosen framework should help you digitize your outputs from your security tech stack and bring 'em into generally any tool that you have out there to give you a single pane of glass into compliance. But, um, you know, it's still subjective, right?

So you have a subjective, um, deliverable, right? That, that somebody has to look at and agree with. Then you have largely manual processes.

So why would they spend the money to automate them if we can't even agree on a standard, uh, to report against them? So you see, you see NIST and, uh, FedRAMP is moving towards more of the compliances code NoCal, but I think we, we lose track of what compliances code is. The digital outputs, the machine readability of these exports saves a lot of time.

But also, you know, compliance is code the digital readability. And your CICD pipelines also operationalizes your security, right? In my pipeline, I'm making sure that I'm building in security.

Then I make sure I build in, in compliance on top of security. And then at the end of that, I had GRC outcomes at the earliest onset because it is just too expensive to try to fix compliance problems or security problems at the tail end of a delivered product. Well, what is the relationship between the folks who are doing GRC and security these days?

'cause sometimes you hear about these two things are melding together, and then the next thing you'll hear is, you know, well, a compliance standard isn't good enough for security, it's just the bare minimum. So, um, how do we meld these functions? Or should we meld them in the first place?

Well, I, I believe that to answer your, to answer the last part of your question first, I believe you have to, right? I think you, you have to meld them, because if not, you're just gonna burn time and money, which no, CO has enough of, right? Their people are doing a thousand things and nobody steals work, right?

They just keep stacking it on top of it. So they're just throwing more trash in on top of a dumpster fire and not putting it out. Um, I think what we need to do is improve, um, is look at, you know, your security tech stack should give you GRC outcomes, compliant outcomes, right?

You should be able to rely on the fact that my scanner is doing all of these things that needs to be done. And at the end of it, I have good reliable compliance because, Michael, I agree. Compliance does not equal security, but compliance was designed right to enforce security, and it should have been a roadmap on how to be more secure.

It's become subjected over time. It's, it's developed cascading problems, different verticals viewed risk management and compliance differently. And all of that has been, like I said before, just increasing the dumpster fire.

So I think we need to kind of take a breath, right? Recognize where I can get my GRC outcomes. I can take my scanner through the use of, uh, uh, you know, integrations, bring in my scan data, review that, scan data, digitize the compliance process, put it into a single with pane of glass so that not only CIS os understand the state of compliance in, in their environment, in their enterprise environment, they also understand the state of security.

So rather than having two siloed entities, which compliance people, security people, they did, I mean, they getting them to talk in any major organization is next to it possible. Um, that's just two of the biggest silos in this. And then you stack legal on top of that, or third party risk management or all of the other things that are coming about, uh, with CMMC and the, uh, you know, GDPR and all of the different, uh, cybersecurity frameworks that we have.

You just have a recipe for disaster. So you've gotta line that up. The only way that, that any compliance person is gonna line that up is by working with security practitioners, working with their operations team, developing what, developing what good operational security looks like, and then, um, getting compliance outcomes at the tail end of it.

And not using compliance as just a checklist, because that's not what it was intended for. That's just what it's turned into. So, as you kind of noodle this a little bit, I always felt like part of the exercise was to try to get those auditors in and out of the building as fast as possible.

And, but it seems like if the processes themselves are manual and the data's everywhere, doesn't that just increase the total cost of the audit? I, I believe so, right? So, uh, I believe that if you're, if you're not, you know, putting evidence collection on a timeline, this, here's what's happening.

You set up a continuous monitoring plan manually, you're assessing controls if they get assessed in accordance with the schedule, right? It, it, it's built into a three year process. And then six months before the audit, people are scrambling around, right?

Then the auditors have gotta find the data, get the data, review the data. Um, the CCC M process should enable you, uh, if executed properly to have all that evidence audit ready all the time, right? Single pane of glass.

I know people laugh when you say that, but I'm a thorough believer in putting everything in one place where you can see it, right? And then putting a schedule over it so that it's pulled automatically, right? You don't have to reach out and wait for a week for somebody to open their email in order for you to get your evidence requirements.

It's already there. Oh, this is out of, uh, this is out of freshness, right? That you get a notification, it says, Hey, uh, you need to, you need to provide this evidence.

And it shouldn't take a compliance person, right? Setting an email or walking across the office to say, Hey, I need this scan, or, Hey, I need you to update this control implementation statement. That stuff should all be done, uh, through automatic notification.

And where you can, particularly on the technical controls, provide automated updates right there. It's really easy to update. Um, um, you know, your RAC data, it's easy to update any of the access controls.

It's easy to update, uh, any of the scanned data, right? RA five should be, uh, entirely automated if you're working into a FedRAMP or, uh, uh, in this 800 environment. There's just, I could go on and on.

There's just so many places that you can minimize the pain points of an auditor by bringing that data in. It's there, it's, it's in a readable format that they're ready to go check, check, check down their checklist, and then they're out the door. So I believe it should, uh, and, and, and speaking with many three PAOs, they agree, right?

They don't want to be there, right? They, they don't. They don't, they, and no three pao that I've ever talked to wants to fail, somebody to audit.

They want you to have a good plan. And in order to have a good plan, if you have a good automation with good evidence collection and good policy reviews and workflows, that can all be optimized into a single, single pane of glass, which gives CISOs increased visibility and allows them to make decisions in near real time vice scrambling to prepare for an audit at the last second. Is this a psychological problem?

And I ask it from this perspective, it feels like we historically, always thought of as an audit was an event that was gonna occur at a given time period. And therefore we would get organized for the audit when it was coming. But it's really should be a process, right?

I mean, it should be something that is just continuously running, and it's not really an event. I, I totally agree. I I believe it should be something that is planned for, you know, you're gonna get an audit.

It shouldn't be a surprise, right? Even a surprise audit, you should be audit ready, right? Through the use of the CCM practices.

You should be audit ready all the time, regardless of, uh, that should be platform and not agnostic. That should be everybody's goal to be ready for an audit 24 7, 365, and to have everything in, in a workflow that you can produce evidence on, on demand. It's unfortunately, right?

You get stuck in these silos where people are comfortable as the value of change is, is not recognized, uh, compared to the, the value of, of not changing, right? So if you have these processes, they've been in place for a long time, people are comfortable, right? You know, you know, every every company has a compliance guy that's been doing it for 30 years and has done it the same way for 30 years.

So, quite frankly, automation scares 'em. And then add AI into the automation, right? That's the latest evil word, right?

So instead of recognizing that machine learning can actually shorten your time by about 80%, if not more, right? It's looked at as either one, you're gonna lose people, which nobody wants to do, or you're gonna lose control of processes, which these things are not true. Every compliance analyst is overworked, every auditor is overworked.

There's way too much work in order to meet the compliance standards. Uh, and without automation, without the use of ai, right? I, I think we're just burning capital that no CIO has, And the audit itself, at least in my experience, is only true or current for approximately 10 minutes, because then somebody changed something and, and then, you know, we're back to the same cycle all over again where we're outta compliance and maybe there'll be a surprise audit or whatever, but to your point, did we just inject too much stress in this system and we're kind of our own worst enemies?

I, I believe that, that that part of it is there, right? I think part of your statement is true. I think that the other contributing factor there is that technology outpaces compliance, right?

And compliance struggles to keep up, right? So I, I believe that compliance is doing the best job it can. I believe that, that every AO is doing the best job that they can, uh, in light of the current situation.

But let's face it, the drivers for technology, right? The drivers for business practice against technology are going to supersede compliance, right? Because if you're telling a company that they have to, that they have to make money, right?

Or they have to be compliant, right? But very rarely the two line up, um, and where you, they can be accomplished ea easily, the company's just gonna fault on making money, right? So we as a, uh, we as security practitioners and compliance practitioners had got to find better ways to make compliance, keep pace with technology to improve business outcomes.

And right now, right? Without the use of any type of compliances, code automation, um, in, in industry without the, uh, it's being used, but it's being used sparingly. So without that, that being put into place, you know, getting the good business outcomes and being compliant, it's very difficult.

And the paradox here is that every nickel we spend on compliance is one nickel that we don't have to either bring back to the bottom line or invest in something else, right? Correct. Right?

And so, I mean, I think it was 46% of the companies, uh, did, or 36% of the companies, and I'm working off the top of my head here, didn't recognize compliance as a, as a business value. And then 46 of them just didn't have the money. So, small companies, right, aren't gonna have the money to stay compliant all the time without automation, which we've seen in the report.

Um, and most of the companies that are embracing automation are small companies that need to be compliant nor to get into the marketplace, uh, and to make real money. So it is, it is like a snake eating its own tail, and it be becomes really ugly. All right?

So what's your best advice to folks then? I mean, what's that one thing that you just keep looking at and makes you shaking your head and go, folks, we need to be better than this Guys. It's not hard.

It's not hard. Find the places that hurt and automate them, right? And everybody has the same pain points if you sit and talk to 'em, right?

You, you have a limited amount of experts or ses that operate in compliance and security. You have the, a limited ability to stay on top of compliance frameworks and the regulatory changes that happen. And you have a limited, uh, you have a finite amount of capital, you have a finite amount of money that you could throw at any problem.

So find your top five paying points, right? And, and do some research on how to automate 'em. And I think everybody's gonna, gonna come to the same place, automatically collect my vulnerability data, write it automatically to an issues tracking, whether it's a poem or a workflow, or ServiceNow or a Jira manage that automatically as it's fixed, my compliance is updated.

I always have good dynamic operational control assurance, and in a single pane of glass, I can see what's going on all the time, right? And this is not hard. A lot of people have done this right on the spreadsheet by writing scripts.

So now all of the, there's a a handful of platforms out there that'll take it to automate this for you. Second, become familiar with what, uh, compliances code is, right? Get familiar with Osco.

There's a lot of resources on the NIST website. There's A-O-O-S-C-F is, there's got a lot of good resources, right? Let's be, be familiar with what I'll call, you know, compliance benchmarking, where, um, these, instead of using compliance as a check mark, use the system on the front end to achieve appliance outcomes.

Um, and, and I do that by doing a business impact analysis, right? Recognize where compliance actually will improve your business. Because I think everybody, if they looked at it, honestly, would see that there's probably four or five spots that they could automate and save a bunch of money, and then take your time and incrementally deploy it, and then go back and do metrics, do a metrics assessment.

'cause if you're not measuring it, it's not, it's not successful. So figure out how much time does my guy do it in Excel? And then how much time do I do it after I automate work slow, right?

It, it's, it's, it's, change is painful on everybody and it's scary, right? So don't, don't boil the ocean, right? Chop the elephant up in the small pieces and eat it a small piece at a time.

Pick your number one control that you think is the most painful, right? To me, it's RA five, right? Get your security scans in, automate the outputs, generate automated workflows to repair the issues, and then update your compliance and see where you stand at the end of that.

And I think you're gonna find that you saved a bunch of money on your security team, your compliance team, and your operations team, all of which has to touch that workflow at one point or another. All right, folks, you heard and hear words to live by it. If it's not fun, go find a machine to do it.

Hey Dale, thanks for being on the show, Michael. I appreciate the invite. If anybody has any questions, happy to answer 'em.

All right, back to you guys in the studio. com is the leading resource for news analysis and education on challenges facing the cybersecurity industry. com covers all aspects of cybersecurity, including data security, DevSecOps, cloud security, application security, network security, security threats, and more.

com has the largest selection of security content featuring breaking news, blog posts, podcasts, and more. com to learn more. com.

Home of security bloggers network. Hello and welcome to the latest edition of the Techstrong AI video series Army host Mike Bazaar. Today we're with David Brola, who's technical director for NCC group, and we're talking about AgTech AI and some of the security misconceptions that go with this next wave of AI technology that seems to be taking us all by storm.

David, welcome to show. Thank you so much. Glad to be here.

I'm not even sure that we all understand exactly what an AI agent is these days, but there are gonna be security implications. So walk us through what's going on here and, and how do we think about this in a way that maybe we can get in front of something for a change? Yeah, for sure.

Well, as we all know, AI has been the hot topic in technology and security for the past three or four years now. And AG agentic seems to be the buzzword of 2025. And there are a lot of misconceptions surrounding what AI is or what being an agent means.

But overall, we're talking about a agentic ai. We're talking about the same systems that we've had for the past three years in terms of like chatbots, language models and so on. But the differences that we're equipping them with the ability to execute tasks autonomously on behalf of users.

So for instance, if I had a chat bot and I asked if, hey, I need you to go out and purchase something for me, an agent would be able to come up with a plan of action and execute on that plan to fulfill my request instead of just giving me some sort of response. So what are the security implications of that? I mean, do I need to double check that the AI agent is behaving as intended, or can somebody hack into that and just take it over?

Yeah, great question and one that I don't think enough organizations are really looking at right now. So at NCC group, we do security consulting, and we've had the opportunity to be exposed to dozens and dozens of different implementations of AI agents. And one of the trends we've observed is that the more functionality and power you give to your agent to execute tasks, the more dangerous it becomes from a security vulnerability perspective.

'cause all of a sudden the impact of compromising an agent where before when it was a chat bot was limited to just giving the user some sort of negative response. Maybe you cuss them out or, or do something else that you wouldn't want it to do. Uh, but now you can actually cause them to execute a task on behalf of a user.

So I might be able to purchase products, I might be able to post, I might be able to manage accounts. And what organizations aren't thinking of a lot of the time is how do we properly manage access controls when these are still statistical systems? I mean, when it comes to traditional technology, unless some random cosmic flare changes the way that my program behaves, it's going to do the exact same thing every single time.

But with AI agents, if I as a threat actor am able to get some form of input into its context window, maybe it read something from my account, read something I put online, then I have the ability to manipulate its output and cause it to run tasks that it otherwise wouldn't have performed. If that's the case, then should I have a lot of smaller narrowly defined AI agents to make sure that the blast radius is limited? Or do I have one kind of Uber AI agent that is just extremely well protected?

Yeah, I think that when it comes to ai, we have to change the security model that we approach it. Having a monolith where we try to make everything surrounding the singular agent secure, uh, hasn't seemed to have panned out for organizations that have tried it. In fact, your suggestion is one of the most successful that I have seen, and it's a technique that I call dynamic capability shifting, where when an agent is executing on a task, when it needs to acquire data to run that task, its permission level is dropped so that the only thing it can execute is the narrowly defined set of capabilities it needs to fulfill the user's request.

So to, to give an example, let's say that I have an Amazon purchase bot and I ask it, Hey, I need to know the latest reviews for some product that I'm interested in purchasing. If I go and ask it to summarize those reviews, it trusts what I tell it to do because I'm the user. But it knows that it can't trust the reviews themselves.

So what the agent should do is the backend code, or, well, first the agent should recognize that it needs to perform a summarization, which operates on untrusted data, tells the backend code that it's going to summarize the backend code, then drops the level of permission that the AI agent has to execute tasks as such that it can only summarize content. Then the agent gets that content, performs the summary and returns the response to the user. That way, if I compromise the agent, it has nothing that it can do anyway with my account because it's permissions have been dropped.

Aren't we gonna have the same problem then with the AI agents that we've had with humans where basically they're overprivileged and then people start escalating those privileges when they hack into them, and then all hell breaks loose. So is this just a replay at a different level of scale? Yeah, that's a great observation and it's really what I like to call the intern in the middle scenario.

AI agents act much less like traditional software code and much more like an untrusted overconfident intern in our application architecture. So in the same way, if you wouldn't trust, and I apologize to any interns out there who are great, I'm sure you're wonderful, but, uh, if you as an organization wouldn't trust your intern with the keys to the kingdom, you wouldn't let them make purchase decisions, uh, wouldn't let them interface with highly important customers, you also shouldn't allow your agent to perform analogous tasks. So in the same way that if I, as a manager told my intern, I need you to retrieve an invoice from a customer that I don't trust, uh, the intern probably shouldn't be facing that untrusted input.

Instead, they should go through a trusted middleman, like a comptroller who's behind the scenes handling all of those trusted operations. com/research that goes into architectures on how you can dynamically limit tho those kinds of, um, trusts that you give to models. And really organizations that aren't thinking about this in terms of how we can trust data dynamically are going to quickly miss the mark.

So ultimately, who should be in charge of securing these AI agents? Is it the security team or is it the business and analyst folks that create them that know what it is they're supposed to be doing? Ultimately, I think that we need to look at shifting security back with traditional code.

We often have a point and patch perspective, or we point out a flaw or a chain of flaws, patch them, and then we're good to go. That that vulnerability is not there anymore. But with AI agents, we need to bake security into the design at the architectural phase of the application.

So I need to construct the way that my components interact with each other in such a way that the agent is untrusted by design or at the very least, limited as much as possible. So in other words, I can't have like some sort of master monolithic agent that's running all of this and then later figure out, oh, well this is a problem and patch, you know, this prompt injection here or this piece of data that it gets there that is just untenable in the long run. Instead, we need to be looking at agents from a dynamic perspective such that whenever we're designing our application data flows, we see the agent receives a piece of untrusted data in this operation, therefore we need to change what it can do to match what that data needs.

How long do you think it will be before we see an AI agent get hacked? A lot of folks saying security, if you can imagine, it's probably already been done, but, um, where are we on this curve? Yeah, great question.

When it comes to proof of concepts, we're already seeing these in the wild from bug bounty hunters who have dug into some of these production applications by major organizations found flaws and reported them, and typically they're patched within a reasonable timeframe. But I anticipate that it won't be long before we start seeing prominent threat actors use these for truly malicious opportunities rather than just as an example to show their skills. So it wouldn't surprise me if within the next year we get our first major AI hack in the news.

And These hacks, as far as I understand it, the AI agents themselves are trained using LLMs, but they're built using traditional coding tools and all the dependencies that go with that. So if there's a flaw deep in some coding tool somewhere, want the bad guys just use that to leverage their way up into essentially commandeering the AI agent. Yeah, and that's a great example of a supply chain attack that we've already seen be executed out in the wild.

To give you an example, on hugging face, they have countless repositories of different AI models that you can download. And so even if the model that I download itself is trustworthy, quote unquote in the sense that it operates at a decent level of performance, there's nothing stopping, well asterisk, nothing stopping that threat actor from embedding malware into the code used to run the model itself. So I could do something like give you a model whose weights contain what is known in in Python as, uh, pickled code, and as soon as you download this model and run it on your system, it will give me code execution and I've compromised your environment.

And so there are a lot of different ways that we've seen threat actors compromise AI systems, not from within the AI itself, but the software used to run the ai. And that is easily as dangerous as any, as any flaw of vulnerability within the AI software itself. Well, you think there'll be regulations emerging around these AI agents as it pertains to security and governance, or do the people in Congress really understand what the, the, the level of conversation really is?

Yeah, that's an interesting discussion to be had. I think that both the Biden and the Trump administrations have proven themselves to be fairly, uh, bullish when it comes to ai. But in terms of security, I don't think that the industry itself really knows the direction that it needs to go much less those in our government who are making these legislative decisions.

Ultimately, the message that I like to give people is that security when it comes to AI has not rewritten the fundamentals. We're still dealing with the same types of security issues that we've seen for the past 30 or 40 years. Just the nature of how those are executed upon have changed in AI environments.

So as a result, I think we need to be thinking less about what specific AI regulations do we need to have that account for AI specific vulnerabilities, but rather what legislation do we need to bring AI systems up to the same security standard that we've seen for other technologies in the past? And How many AI agents are there likely to be in need of some level of security and governance? 'cause in my mind, I can imagine that I might have 10 for various tasks, and then if each person has 10, well, you know, you're getting into the thousands pretty quickly.

Yeah, I mean, ultimately I think that every agent needs to be designed with security in mind. And right now when we do these types of assessments, I would say less than 5% have been structured with the requisite security considerations and guarantees that they need to have to be secure whenever they're deployed out into the wild. So what's your best advice to folks?

Because there's a lot of security people that struggle with the same issue all the time, is the tech gets ahead, uh, people don't think through the security implications. How do I get, insert myself, I guess, into that conversation in a way where I can have an impact now versus trying to clean it up a year from now? Yeah, great question.

When it comes to ai, we need to change the way that we think about trust from just a component level perspective to one that incorporates the data itself that these systems are using. So in other words, it like with a traditional application, I might say that as long as my dependencies are secure, and as long as my developer's code is secure, then I don't need to worry about security vulnerabilities cropping up outside of maybe business logic. But when it comes to ai, in a sense, everything is business logic.

So if I misplace an assumption about how I trust my ai when it receives untrusted content, then I am going to introduce security vulnerabilities into my system because that AI is no longer a trustworthy component within my environment. So we need to think of trust being polluted downstream. So whenever an AI receives a piece of untrusted input, I need to consider the AI exactly as trusted as the input that it receives.

So in the same way that I wouldn't trust an Amazon review to make changes to my user's account, I shouldn't trust an AI agent who reads Amazon reviews to make those same changes to my account as well. Do you think that the bad guys are kind of sitting back and maybe having a little chuckle amongst themselves? 'cause essentially we are now exponentially about to increase the attack surface, and they're like, and it's even better, they don't even have any idea what the security issues are.

Yeah, well, somebody who does security consulting on the offensive side where my job is to look for security vulnerabilities, I think all of us get a little bit of a rush when some new grand technology comes out that takes the world by storm. Because the first thing on our minds is, okay, what's next? What can I go out and hunt?

But in the same way that we saw the same pattern with cloud, with blockchain, with internet of things, we're going to see the same cycle of the industry rushing to implement a new technology, doing it incorrectly, seeing a lot of security vulnerabilities, and then finally having those vulnerabilities patched by new security standards within the industry. So in the same way that I think threat actors probably are, to your point, really excited to see a new technology take the world by storm, I don't think that it's going to be something that inevitably cripples us. I I do think that this is another part of the same cycle that we see every five years or so.

Well, let me ask you this then. Will it play out this way? Someone will create an AI agent and another person will create an AI agent for managing the security of that agent and for governing it, and the agents will kind of eventually, um, circle each other in a way that gives us better security because there are checks and bells.

Yeah. So almost, uh, turtles all the way down, or an agent or a borrow, Something like that. Yeah, It's interesting.

I see a lot of organizations trying to approach it from that perspective. And I think that what we observe is that the more agents you put in in a chain, the more likely it is that one of 'em is going to pick up on the fact that, wait a second, something's going weird. Now we should probably shut this down.

But in the end, AI itself, like I always say, is a statistical model. All we can do is strongly suggest that it follow the rules, and it just takes one threat actor to suggest even harder than our developers to get it to misbehave. So as a result, I don't recommend relying on AI to be the security control that protects ai.

I, that's what I would call a soft control. It's a defense in depth measure. Rather, we should putting, be putting in and implementing hard security controls from an architectural perspective such that even if the AI wanted to misbehave, it would have no opportunity to because of the code surrounding it.

Right? Folks, you're heard it here. No matter how you look at it, no matter how awesome AI is, the more there is of anything, the harder it's gonna be to secure.

So AI agents are no different. David, thanks for being on the show. Thanks so much.

Enjoyed it. All right. And thank you for all watching the latest episode of the Techstrong AI video series.

You can find this episode and others on our website. We invite you to check them all out. Until then, we'll see you next time.

Good day everyone, and welcome to the 5G Factor. I'm Ron Westfall, research director here at the FU and Group, and today it's a good one. I'm joined by my esteemed colleague, a literally a Blanchard, a fellow research director and practice lead for our AI devices practice here at the Futurum Group.

In fact, Olivier just recently returned from the Samsung unpacked event. And with that, I think we have some good insights on what's going on in the 5G and mobile ecosystem realms. And with that also we'll be focusing on just that the things that have really jumped out of recent, you know, vintage, uh, that merit our attention.

And so with that, Olivier, welcome back to the 5G Factor. How have you bear been bearing up since the last time you've been on the 5G Factor? I can't even remember the last time I was on 5G Factor.

I've been, I've been dodging your invites for a while, not because I wanted to. I've just been on the road it seems like a lot more in the last six months than I have ever. Um, so I've been, I've been doing well.

It's, you know, it's, uh, it's hard to keep track and, and to keep pace with, uh, all the changes in the industry and the hurricanes and, and all of the, everything that's happening all at once. So, uh, yeah, it's good to have a minute to to Be back back and, and catch up with you. Yes, indeed.

And you know, perfectly understandable and right, you are. It's been a very kinetic last, you know, well year plus really overall. And, uh, and I think our first topic that we're kicking off has played a major role.

Yes, AI again, however, this time, this is something that I think is really galvanizing, you know, attention. And hopefully by the time our recording is published, it'll still be somewhat relevant because matters are moving very quickly. And that is, uh, just on January 20th, uh, China-based Deep Sea released R one.

It's a reasoning model that basically is outperforming, you know, open AI's latest, uh, oh one model. And that is, uh, verified in, you know, various third party tests. And so what is interesting here is apparently 150 researchers at a Chinese hedge fund, also known as Deep Seek, has out flaked the entire work of tens of thousands of engineers and basically almost the entire Western scientific community.

And what is this just with a handful of modified Nvidia, uh, H 800 GPUs? Well, let's see, let's stay tuned on that. Uh, more likely at least, uh, from my perspective is that Deep Seq was trained on more than 50,000 H 100 GPUs.

And while that is good, it doesn't automatically mean that the AO AI ecosystem apparently has reached a point that can train on materially less infrastructure. And that means we still, despite this breakthrough, we'll probably need to continue our massive commitment to figuring out, uh, ways to optimize AI training as well as AI inferencing. And that is a, a naturally linked to at attaining the main objective of artificial general intelligence or a GI.

Now, a little more background, uh, there's a lot of threads here, but one I'm gonna focus on initially here to kick off the conversation is that according to Alexander Wang, the CEO of scale ai, he's the one that basically I think initiated the notion that, okay, what is going on here is that the Chinese have access Nvidia advanced GPUs on a wider scale than many people realize. And the reality is that the Chinese labs, that they have more H one hundreds that many people think, and he added and shared that understanding, is that deep Seek has about 50,000 H one hundreds, and they can really, uh, uh, can't talk about it because it's against the export restrictions that the US has put in place. And as a result, they have many more ships than, uh, folks, uh, really, uh, I fully understand.

But that's changed, uh, obviously dramatically. And Elon Musk, uh, also seconded the motion. Now, Nvidia, on the other hand, has insisted that deep seek is an excellent AI advancement and a purpose example of test time scaling.

So this is an important technique now that will, I think, get a lot more attention because of what, uh, Deepsea has basically achieved. And, uh, in addition, uh, deep seeks work illustrates how new models can be created using this technique, leveraging widely available models and compute that is fully export control compliant, uh, emphasized nvidia. And so, uh, this means that infrastructure requirements, uh, that require significant NVIDIA GPUs and high performance networking will still be needed into the foreseeable future.

And this aligns with what they see as that there's now three scaling laws, pre-training, post-training, and now test time scaling, uh, the newest one on the block. So in effect, uh, Nvidia is adjusting that the deep seek used export compliant equivalents of an A MC gremlin and pacer and soup them up to the top line equivalence of Maseratis and Bugattis. I know it's a loose analogy, but it's like, wow, this is a pretty remarkable that if they only use export compliant GPUs to achieve this.

Well, I think there's a lot of counter append out there, and, uh, as a result, there's this mystery, like what are is the GPU cluster that deepsea actually used? I don't think you're really gonna know for a little while, but this is definitely, I think, fueling all the speculation just how replicable is this? Just how cost efficient is this really?

But the outcome is like, okay, great. Now we have a AI training that could be done on a much more cost effective, much more energy efficient level then yes, and it's also open source space. It's good news for the rest, the AI ecosystem.

Uh, now officially, uh, Deepsea claims have used only about 2048 Nvidia H 800 chips, which, uh, to train the R one model. And that is was alongside the, uh, 10,000 older generation, a 100 GPUs that had already attained, uh, before the US uh, imposed export controls. Now all this, uh, I think is going toward NVIDIA's credibility on US export policy.

Now, naturally, NVIDIA has to claim that yes, that the, uh, chips that were used were export compliant, and that's understandable. However, I think as we dig deeper, and if we take, you know, the opinion of, you know, folks like the CEO of signal ai, uh, then I think that means that there's more going on here that can be officially disclosed. Bottom line is that, you know, many of the actions that are backed by China's government, I think as we understand, you know, we've seen this with the Huawei for example, is that it constitutes what can be characterized as a giant psyop.

And it's, in this case, Han is attempting to create a little bit more uncertainty among the investment and tech community about AI's prospects in terms of it being, you know, primarily driven by, uh, US or Western technology knowhow. And as you can see, it did effectively shift the spotlight from, you know, the half trillion dollar project star, uh, uh, uh, um, uh, uh, which has, uh, its own set of, uh, major, uh, questions. And so, uh, we have been, you know, really looking at, you know, how can we improve scaling laws and seeking efficiency with AI despite the deep seek breakthrough.

And, you know, uh, again, we're been using open source, uh, rag fine tuning and forking capabilities, all these AI techniques to allow smaller models to be more performant. And we've already seen companies such as IBM, meta minstrel and others have stepped up and really have made, uh, I think an impact on how this can be better achieved. That is, you know, right size, large language models that are better performant.

And, uh, with that, uh, Olivier, from your perspective, I know you've already, uh, provided some, I think, very valuable insight on this. And what is your take on what's going on here? I know this is just one aspect here, but what else do you see going on here?

Right. Well, I, that, that, that was a lot. Um, I feel like the, the meme, you know, with a guy like this, and it's like he's got the, the board with all the red strings just kind of connecting everything.

Um, so I think that everything that you said is, is, is on points, uh, including the, uh, the, the, the, the theory or the hypothesis around Chinese ops. Although I, I would caution that, um, if, if we equate the AI race between US and China to any other arms race or space race in the past, the, the, the, the prevailing sort of tactic to, um, harm the other side, your opponent isn't to necessarily undermine credibility, uh, of their model. It's to get them to spend a lot more resources chasing the same thing instead of trying to make them efficient.

So by, by releasing deep seek, um, China, assuming that the Chinese government is involved in some way in this strategy, in the timing and in the, uh, the, the tone and tenor of this release, um, the, the, the objective for China should be to just let the US overspent on infrastructure and AI resources instead of taking this sort of more efficient scrappier approach that they took to achieve the same results with fewer resources. So I, I, I take the, the notion that that China is trying to undermine US markets and US investments in AI with a grain of salt a little bit. Um, but having said that, I, I think, I think most people are actually missing the point with deeps six.

I don't, I, I don't, I don't know that we're, we'll still be talking about deep seek in six months. I think that the, the larger trend, which is something that we've been talking about for the better part of the last year, is that there's been an evolution. There's been a, a trend line in, uh, in AI training and AI inference that transcends deep sea, uh, and, and any other company like it that is very likely to pop up in the next, you know, six to 18 months.

And it's this, the, we're we're at a confluence of, uh, an increased and, and sort of systemic, um, improvements in the efficiency of the model on the training side, on the inference as well, but definitely on training, um, where, um, models that, that had to be trained in the cloud with massive resources behind them, massive inputs of power of chip and compute to, to train these models have become so much more efficient in the last year that what used to be only trainable on the cloud can now be trained on PCs and in some cases on mobile. So we're seeing these, these models sort of shrink and get, get faster and cheaper to train, uh, just by virtue of the fact that they're, they're, they're becoming more efficient. On the other hand, we have this, this other element, which is the chips are getting more high performance as well, which is the whole value proposition between NVIDIA's Blackwell, right?

You could use fewer GPUs to, to generate more outputs faster and more cheaply. So there's, there might be a slightly more upfront cost on, on the GPU side, but you need fewer of them to achieve the same results and few less power, less, less, uh, water, all of these other resources, less footprints, uh, data center-wide. So we have these two efficiencies basically working together to make AI training a lot cheaper and a lot faster.

This was happening already. It is gonna continue to happen. Uh, it's, it's one of the things that is fueling the proliferation of AI processors and AI models from the cloud to the edge, and creating this more hybrid ecosystem of basically sort of like cloud to endpoint, uh, AI training and, and inference, uh, which, which I think is the reality that we'll be existing in, in, in, uh, as, as early as later this year.

Uh, and we'll touch on that again when we get back to our coverage of the, uh, Samsung event last week, because there's actually an element that that plays into this. And so, um, the issue with, uh, with, um, stargates and the enormous investment numbers that we were talking about a week ago, and sort of like this injection of excitements in, um, in AI infrastructure specifically in the United States, seemed a little bit, um, I don't know, warts last week when the announcement was made, the, the impression that I had given what we just talked about and, and sort of validated by the deep seek announcement, uh, a few days ago. And the, the sell off, uh, of, of US tech stocks in, uh, in the last, you know, 72 hours, um, is this, um, this is already happening.

And, um, it, it's the, we don't need to spend trillions of dollars building massive data centers that are going to house millions of chips and, and, and servers and racks and, and require nuclear power plants and, and some kind of, you know, weird re-engineering of our water supply and our water management systems. We don't, we don't need to, to completely just stop everything that we're doing and build these massive projects, um, because the models and the chips are becoming more efficient and this federated distributed AI training and inference is already happening and the costs are getting lower. So I think that what we're seeing last week is, um, let's go back two weeks ago, two weeks ago at CES Jensen, um, introduced sort of like his vision for the next phase of what NVIDIA is about.

And we're gonna talk about Nvidia because NVIDIA has sort of like the, the market power and the best position in terms of, of GPU infrastructure, um, and, and IP when it comes to training, um, AI at scale. Um, his model at CES, the introduction of Blackwell, all the things that he wanted to do, what he talked about in terms of, uh, training for automotive, training for industries, doing all these, these virtual sort of, um, you know, digital doubles and digital twins of the world and, and different environments is still valid. None, none of that has changed.

Um, but, um, it, it's the 2025 spend on this. And the reality of, of where technology is today in 2025 with 2025 chips and 25 models, the the entire 2025 current layer of, of AI solutions and AI IP is absolutely not where we will be in 2028 and in 23. And I think that Stargates misses the point in that it, it looks like, um, the type of spend that we would put together if we assumed that these efficiency advancements, these efficiency improvements were to stop this year, right?

Uh, so we're looking at bills for 2030 with budgets that reflects where we are today in 2025, not where we will be in 20 28, 20 30, 20 35. And so I think that they're, they're a little bit bloated. I think it's, it's warped expectations.

Um, I think the front end investments, uh, may be accurate. It, it may be the right number, uh, dollar wise, but I don't think it, it's, it's necessarily, I think it may be over-indexed on the data center side, is what I'm saying. I think that, um, taking a more holistic approach to the entire ecosystem of, uh, of research of chips all up and down the, uh, the, the AI value chain from basically wearables all the way up to the data center is, is a more realistic approach.

So I think that, that we're gonna end up spending a lot less on data centers and a lot more on production, on supply chain and on all of those middle layers to, to create a, a more sort of hybrid, uh, ubiquitous AI ecosystem. Um, and, and deep six's announcements. And obviously the reaction to the market, which I think was an overreaction, but that's a whole other story, is just one of several proof points along the way that the increasing the rapidly increasing efficiency of, of AI training and AI infrareds workloads for that matter, uh, and the, the, the sort of like diminishing curve of costs associated with that is, uh, is going to be a disruptive force in some of the calculations that we've had about how many chips we're gonna need to be able to, to, to power this new, um, ai uh, AI economy, right?

Um, and that's not necessarily a bad thing for Nvidia and for a MD and for everybody else who's involved with this, but it, it, we should perhaps adjust these, uh, these massive numbers that we've been looking at and, and look at it more as a term in terms of, uh, diversification of chips. So if you're an NVIDIA, for instance, don't just focus, or if you're looking at Nvidia for, uh, you know, in, in your investments, I'm not make giving any investment advice. Uh, so, so, um, don't take that for this, but I, if you're looking only at, at Nvidia for its data center GPUs, you're missing the point.

You need to look at Nvidia all up and down the value chain from, from PCs and devices and iot all the way up to, um, all the way up to the data center. And if you're looking at other companies like, and Qualcomm or Intel or MD or, or, or media tech for instance, uh, you also need to look at those, those device layers and those intermediate layers where there's gonna be a lot of scale to deliver AI organically through this hybrid model, if that makes any sense. There's a lot, Yeah.

And that's, uh, topic warrants that I think we're only gonna scratch the surface, but I think that, uh, our outstanding, uh, viewpoints there that you shared Olivier, uh, you know, first of all, you know, let's look at the big picture here. I agree this is not an existential threat to Nvidia by any stretch of the imagination. And yes, their portfolio has diversified over the last few years.

It's no longer, you know, about, you know, only, uh, GPU, so to speak, or at least, you know, there's that perception. It's, you know, the software and the services, it's definitely been diversified. And yeah, I think, uh, one important takeaway from CES among many was, you know, what, uh, Nvidia is doing with Cosmos that is, you know, taking real world video and applying it to AI training and capabilities so that you have, you know, these outcomes that can be very useful.

And, uh, just that real world settings, you know, how to optimize, uh, say what's going on in the warehouse as an example, and, you know, advancing robotics and so forth. And so, you know, a lot is going on here, but I, I think, uh, what is going on, uh, with a deep seek and, you know, uh, for that matter, uh, project Stargate, uh, I think, uh, that's important to note here is that it's like the thro thing of, you know, the AI hype cycle here. And to your point, I think, you know, okay, what about the half trillion dollars that's been allocated toward, uh, project Stargate?

Uh, what are the implications of, you know, a lot more efficient ai, uh, capabilities throughout the AI ecosystem? And this is a variation of Jevons paradox, that is, if you make something a lot more efficient, that will actually increase more demand for it, because it becomes more affordable and more broadly available. And to your point, yes, when it comes to, uh, hybrid ai, okay, a lot of the heavy lifting, uh, AI trading that's done in its GPU clusters and data centers, you know, throughout, you know, say hyperscaler, uh, networks, uh, a lot of this, uh, can now, uh, uh, be done in a more distributed basis in your edge data centers and so forth.

But also, uh, to your point about devices, yes, you know, the ai, uh, inferencing, let alone some of the training can now be done at the outer edge. And that includes, you know, devices that are powered by Qualcomm arm, uh, media tech and Apple, as well as, you know, across the, you know, the broader edge, you know, uh, including, you know, Broadcom, Marvell, Intel, a MD, you name it, that, okay, this doesn't necessarily mean you have to enlist Nvidia, GPU, uh, GPUs to do this, uh, more distributed edge, uh, training and inferencing, but it's really, I guess, uh, just that, uh, welcome news for the entire a AI ecosystem as well as for the other, uh, chip players out there. And, um, and on that note, let's, uh, now segue to the next topic, which is really AI at, on the device level.

And as we saw at Samsung unpacked, which you were able to join, is that Samsung and announced the, uh, galaxy S 25 Ultra. It's, uh, galaxy S 25 plus and Galaxy S 25, all aimed at setting a new standard for true AI companionship with, uh, you know, basically our, you know, context to where, uh, mobile experiences. And I think that's a, a very poignant way to, you know, position this and market this at the onset.

And what we saw is that Samsung introduced multimodal AI agents, and as a result, the Galaxy S 25 series is really a first step in Samsung's vision to change the way users interact, uh, with their phone, and also, you know, with the, the real world for that matter. And so it's using the, uh, customized Snapdragon eight elite mobile platform to attain a lot of these, uh, uh, goals. And also the Galaxy chip set is, uh, designed to deliver, you know, greater on device processing power for Galaxy AI and, you know, improved camera range and so forth, and let alone galaxy's, uh, you know, uh, pro visual engine.

So I'm gonna stop there because you were there. Uh, you have, I know, in depth, uh, uh, perspective on this. So what were some of your key takeaways from, you know, what Samsung announced with the new Galaxy S 25 Y?

Right. So it was, it was a really interesting event of, of all the, uh, of all the Samsung impact events that, that I could have gone to this, this was probably, uh, one of the more significance. And I think, um, somehow a lot of, of that significance got missed by the coverage of the event.

So let, let me give you my, my take on this and why, why I think it's so important. I've, I've been operating under the, the assumption, uh, as an analyst and as a, a practitioner of, you know, where AI is going and how, uh, I've been, uh, operating under the assumption that ultimately what we want is ubiquitous AI, device agnostic ai, where you walk into a room, it doesn't matter where you are, um, the, the devices that is nearest you and most capable of delivering the agentic AI experience that you're, you're, as a user, you're expecting, uh, and doing it more efficiently is going to be the one to deliver it. So if I'm talking to an assistant and prompting it and saying, uh, I wonder what the weather is today, or, tell me, you know, what are my appointments this morning?

Uh, it doesn't matter if it's my watch, my, my headphones, my smart glasses, my pc, my smart speaker, whatever it is, that's where it's gonna go. And with, with Agen AI specifically, one of the really interesting user experience advantages or value propositions is this hyper, hyper-personalization where your agents learn from you. They learn from your habits, they learn from your needs, from your patterns.

They become sort of like your best friend. They, they know what you want and in what, in what format, and what style and what speed before you know it. But definitely when you prompt it so they can anticipate your needs and, and adjust your environment, your workspace, your calendar, your, your shopping planning, all of it, uh, for you.

Um, however, what that requires is a lot of device to cloud integration. And, and I'm not gonna get into the, the specifics of, you know, how that needs to work from an architectural standpoint and an orchestration standpoint, but it's extremely complicated, and it requires some data to be on device to be cloned in the cloud and to be accessible both in the cloud and on device, uh, so that there's no, there's no lag between, you know, the prompt and the response. So you can have natural, uh, naturally times natural language conversations with, with your agent and your ai, uh, assist it, uh, without having to wait for a response or like, Hey, I'm thinking about it, gimme a second.

Um, what Samsung did though is something very different from that, which is super, super interesting. And at first I thought, okay, this is the wrong approach. And then the more I thought about it, the more I realized, wait a minute, this is actually smart.

What Samsung did is they prioritized personalization and data security, and data integrity and, and safety, uh, by essentially moving a lot of those processes, the training and the inference to the device itself, and blocking it from the cloud. And so essentially what they did is they have two parallel systems. They have, uh, uh, because they're Android devices, there's, there's Gemini, right?

Which is kind of like the Android device to cloud, uh, AI platform that remains untouched. So anything that you do with search, it's gonna go out for Gemini, and, and you have this net and this normal integration, but all of the super personal stuff that it's learning about you, um, essentially sort of like the personality graph that, uh, that an agentic AI is, it needs to be able to build and train itself on in real time based on your needs is, is super personal. And, uh, and Samsung made the decision to keep all of that on the device secure, no contact with the cloud, nobody else is gonna train on your data, nobody's collecting your data.

Google is not collect, collecting that private data. It's on, on your device. Not only that, but it's protected by their NOx security solution and it's post quantum encryption, which means that currently, at least in, in theory, I haven't validated this, uh, but, but based on on what Samsung is telling us, um, it's, it's not decryptable, uh, with quantum computers, which is, which is really nice.

So pros and cons of that. Um, pros, obviously, data security, privates, uh, all of your data remains private. And, and you have this, this, I think, added sense of, um, of privacy that you nor wouldn't normally have with a lot of AI products and, and solutions.

Um, and it's tied directly to Samsung. The, the, the con is that if you lose your device, you have to start all over again with all of that, uh, agent training. Um, but device to device, when you upgrade from the S 25 to the S 26 or whatever the next generation is, you'll be able to do that.

You just can't like, upload your, uh, that private data to the cloud. Um, but all of this is done because the, uh, the chip that they're using, the SOC, the system on ship is, uh, is capable of doing that. And in this case, it happens to be Snap or Qualcomm's Snapdragon eight, uh, elites, which is the, the sort of like flagship platform that, that Qualcomm introduced, uh, at their summits back in October, um, of last year.

So it's, it's the latest and greatest, but it is a custom chip made specifically for Samsung. So it's, it's actually the Snapdragon a eight elites for Galaxy, uh, which has a few additional bells and whistles for, you know, some camera, uh, improvements and also for this, uh, this enhanced, um, uh, agentic AI and device capability. But it's, it sort of illustrates, I think the, um, the, the, the new paradigm of this distributed, uh, agentic AI where sure, you can do, you can train a lot of models and, and do a lot of things in the cloud, and there are things that work best in the cloud, and that should be operating that way just as a cloud service.

But there is also, oh, all right, I'm gonna continue it, it paused for a second. Um, that'll be a nice place to cut. But there's also a, um, a, a, a really huge leap forward in capabilities of these, these ARM-based chips that are on mobile phones, that are on PCs that increasingly are showing up in, in smart watches as well, uh, in smart glasses, uh, in, in essentially every digital product that, that we touch.

Uh, there's more and more capable of, of doing this. And, you know, the, the size of the, the models that you can train and, uh, and, and do inference workloads with on a PC versus mobile versus smart glasses depends a little bit, first of all, on, on the, the system on ship itself, but also on the size and the form factor, right? You can put a lot more processing power in a PC than you can in a phone, and you can put a lot more processing power in a phone than you can with smart glasses or on a smartwatch.

Um, but you also have to think about how all these devices work together and how they can pull their processing resources so that instead of processing something on the watch, you're processing some of it on the watch, some of it on the phone, some of it on the pc. Uh, and a platform like Snapdragon, which is in all of these devices, might be able to just kind of work altogether more efficiently than cross platform, uh, combinations to give a user, um, uh, an enhanced AI experience that is primarily on device or that can sort of separate, um, the needs of, you know, pushing some of the, uh, the workloads to cloud services and then keeping some of them private for a variety of reasons for speed and efficiency, for power efficiency, for cost efficiency, but also for privacy. And so it has implications for, uh, consumers, right?

You and me just wanted to keep things private, uh, and cheap and not having to pay for all of this cloud inference stuff. Um, but also on the commercial side for businesses, because now the more data they can, they can have in-house, the more secure their data might be, um, and the more processed and they can do in-house also, the lower the costs, uh, of, uh, I mean, they don't necessarily have to pay for as many, uh, instances of, of training or workloads that they're pushing out to a cloud through cloud service. They can keep a lot of that stuff inhouse.

Yep. Yeah. I, I think it's good news for say, the health monitoring use case, this, and it aligns.

And that's, I think, one constant theme. Uh, when we saw, you know, with the launch of Project Stargate, Larry Ellison step up and say, Hey, this AI investment has warranted because of advances that could be made in medical research or, you know, tracking medical records and so forth with that privacy respected, built in and so forth. And likewise, you know, just being able to have a device that somebody can have confidence, it's gonna maintain my privacy, but it's also can just do that, you know, detect issues, uh, before, you know, they get outta control or, you know, just across the board any health, you know, benefits, uh, out there.

So that I think is, is certainly the good news. That was one of the, the, the main use cases that, um, I'm not even sure that it was, you know, I think it might've been under indexed a little bit at the announcement itself, but in the pre briefings, uh, we spent a lot of time on, um, on that particular use case, how, uh, Samsung specifically. But I think as a concept, having all of that data and that agentic ai, um, processing and analyzing and recommendation engine on the device as opposed to in the cloud, first of all is more immediate.

But also, um, I think if, if, and I think especially with medical issues, um, I think people want their data to be protected. And we've all been burned so many times with hacks and, and, you know, our, our private data, especially medical data, getting out into the wrong hands or just getting out in the public, this, this takes care of that. And so you have a, a personal assistant that is able to, through the use of different devices and sensors, whether it's a, a Galaxy watch or a ring or, or other sensors, um, e essentially guide you and give you feedback on how well you're sleeping.

For instance, your eating cycles. What your calor can take is, uh, we were talking about things also about a more granular approach to your diet, which you can sort of enter manually, but also, um, the AI knows what you're eating, and so it can sort of extrapolate the types of nutrients that you're getting and not getting. So it's, it's creating and painting a, a, a much more, uh, again, granular and, and detailed and complex picture of your overall health.

And it is intelligent enough to understand your cycles, understand your patterns, understand cause and effects of good and bad behaviors, make recommendations, um, monitor your health. And it's, it's, it's amazing because it's all on the device and it's all secured. And, um, I love this, it, it, it moves some of the, the, I think the, the real true immediate benefits of AgTech AI out of the cloud in areas where it doesn't need to be in the cloud.

Um, so some things need to be in the cloud, some things shouldn't be, and it, it's, we, we now have the capability of, of parsing that, uh, according to our needs and according to our preferences. And that's gonna, that's gonna change that, that changes the equation a little bit for, uh, for how we think about AI investments. And again, it, it speaks to that diversification of, you know, focusing on, on cloud AI training and inference, but also focusing on this, these edge use cases that are becoming much more prevalent and with a, a huge potential for, uh, for adoption.

And the, the footprint. If you look at the mobile industry, if you look at the PC industry, um, even if, if those numbers, uh, essentially the install base doesn't really grow that much, there's a refresh cycle here, uh, that's gonna bring a lot of these AI chips to this, this broad install base. So the market doesn't actually have to grow to show results.

It's the refresh cycle that's gonna show those results and, and push a lot of these, uh, uh, these AI chip numbers out. Um, and that's what I'm hopeful about. Yeah, and I think, uh, that's a great segue for who else can, uh, benefit from, you know, the, uh, the potential large s of the AI ecosystem, uh, overall, uh, growth.

And this ties directly to, well, you know, 5G service providers, uh, certainly, you know, the folks who provide mobile services to, uh, consumers and, uh, businesses, uh, and specifically here in the us. And what, uh, I think is interesting is that we're seeing, you know, major US operators such as at and t and Verizon looking into, okay, how can we play a more integral role and monetize, you know, AI in terms of, you know, uh, being closely linked to the services they provide. It's not, uh, exclusively mobile, but it certainly includes, uh, for example, their fiber services, business services.

But I think what's interesting here is that they have the real estate that we touched on, Olivier, how do we push more AI capabilities, training and inferencing closer, uh, to, you know, where the customer is, you know, bringing the AI to where the data is is certainly the mantra that, uh, we've have heard a great deal about. And so what's interesting is that now we see, uh, that Verizon business has, uh, revealed basically a, a bundle of products that are designed, uh, for enterprises and also cloud providers as well as hyperscalers to, again, deploy those AI workloads at scale by, you know, basically offering a single platform, uh, that is designed to do just that. And what it is, it's called Verizon AI Connect, and it's offering a blend of really the operator's fiber infrastructure along with its power space and cooling capabilities, uh, and with those resources using it, uh, to deliver it, but also it's backed by its virtualized 5G programmable network to really make the AI workloads, um, more, uh, I guess you say, customized to the, uh, customer needs, uh, out there.

And so what's interesting is that already, uh, Google Cloud and Meta have already onboarded meta platforms specifically. And I think that's important because, uh, these are, you know, going to be logically the early adopters of AI infrastructure, uh, platform such as this that are offered by a, a major, uh, service provider. And, uh, not to mention Verizon and Google are also looking at ways to, you know, advance AI services for, you know, things like network maintenance as well as anomaly detection.

So this is in play, uh, this is something that I think will kind of help the, uh, service providers. Okay. Finally, uh, we have a way to use our real estate, uh, to take advantage of, you know, ai.

And this is not going to be necessarily a repeat of things like, you know, mobile edge computing and so forth. There's still that risk, you know, the operators still not, not be able to figure out how can we be integral to this. But I think, uh, this is demonstrating that, uh, they're off to a decent start.

And not to be outdone, a TT recently secured, uh, $850 million from the sale and leaseback of its underuse, uh, c facilities, uh, from, uh, a property company, uh, capital rain as part of its copper retirement plan. And so, uh, this deal closed in January, it involves the asset transfer of 74 properties across the US spanning more than 13 million square feet. So the bottom line here is that the operators are becoming, uh, smarter about how they can take advantage of the real estate assets that they do have today, you know, that is retiring co assets as well as other edge infrastructure.

And, you know, working with, you know, the, uh, the major, uh, AI players out there, that's IE the hyperscalers to, you know, basically come up with mutually beneficial ways to monetize AI in the near future and certainly longer term. And, you know, Libby, what are your thoughts on this? Do you see the service providers really being able to step up and actually, you know, playing a meaningful role in this?

Or is this something that, okay, it's another missed opportunity in the service providers will be reduced to kind of a commodity like provider of the infrastructure for the AI services that are running across, you know, the, the clouds out there? Uh, maybe a little bit of both. Um, so yeah, no, I think, I think it's smart for 'em to do it.

So, you know, the, the situation that we're in is, uh, we're expecting, obviously if we're spending, we're looking to spend half a trillion dollars on building data center, um, infrastructure. It means that we're, we're looking for a lot more computes to come from somewhere. Uh, a a lot of these, you know, you know, builds are, are years down the road.

We, we can't really start yet. So, um, the expectation is that we're gonna need a lot more compute power very quickly. Where can we find it?

And if you have data centers already out there that are underutilized, right? Um, or that can be sort of, you know, a lot of that processing think power can be retested for, uh, you know, higher price or, or more premium services, um, it makes sense for a business to go for that, for that ROI, right, for that opportunity. So one, if, if a lot of those data center resources are underutilized, not used at all, suddenly they can be, you know, assigned to this, if we can charge a little extra for this, because AI is more valuable than, you know, being on a 5G network, um, maybe there's, there's value in that as well.

I think though, that, that you're right, ultimately there's, whether it's successful or not, that that is gonna depend on them and how they package it. And, and it, there are a lot of variables there. I think at some point, those data centers and those resources age out, they're no longer, um, the most efficient, the most cost efficient.

Uh, and, and maybe they just get retired, or the Verizons and at and ts of the world upgrade their systems specifically for those AI workloads. And now we start seeing a, a transition of spend where they also become, you know, they're, they're buying the black well powered, uh, racks and, and, and they're becoming more AI focused, and they become part of this sort of like, uh, ecosystem of, uh, of, of AI workload, uh, training and inference services, and it's all interwoven. Maybe that's, that's possible, but for right now, at least for the next few years, while we wait for all of these, these massive data center builds, uh, they're there with capacity.

And so absolutely they should go after that market and, and see what they can make of it. Um, worst case scenario, it doesn't work at all. Middle best case scenario, uh, they make some money, it becomes commoditized, and eventually it just kind of dies out because, uh, they're outperformed by other outfits.

And best case scenario, they build a whole new business model that's, uh, that's gonna be really lucrative for them. So it's worse than the shot. Yes.

Yeah. And I think, uh, this is, again, you know, AI and it's open-ended possibilities. Uh, yeah, there's just a myriad of variables here.

So on the one hand, it's good news for those se uh, service providers. Certainly, uh, test time scaling, uh, is, is introduced to the possibility of, okay, AI as we do it could definitely become more commoditized, as you pointed out. And thus, you know, that's good news for the ser uh, service providers.

It just become a lot less expensive to, you know, do the, uh, AI infrastructure hosting and so forth. Uh, now they, however, is, uh, will this result in monetize outcomes for them? And so, yeah, that is again, that kind of, uh, one of the many, uh, uh, uh, I would say the three doors that you presented, uh, possibilities.

And it's hard to bet right now on the operators because the entire a i ecosystem is basically going through a lot of flux as we speak. And so, uh, we'll, uh, we'll come back to this. We'll talk more about it.

Yeah. We should, we should come back to this six months from now and see, uh, Definitely. Or let alone yeah, say after Mobile World Congress, uh, you know, a lot of changes from that.

Um, so this has been great. Uh, thank you so much Olivier, for coming on board, and again, appreciate, uh, the, uh, opportunity for you to share your thoughts. Yeah, thanks for, thanks for having me on.

And I am not going to Mobile World Congress as of now, this year. I'm skipping it. I'll probably go next year.

Um, but, uh, but I'll be, I'll be looking forward to, uh, to the announcements surrounding MWC. 'cause I'm, I'm sure this, this very topic is gonna be, uh, is gonna be one of the major, uh, themes of, uh, of the trade show Yeah, Yeah. That we can bet on.

And that's something I think we kind all agree on. A I might as well call it AI World Congress for the time being. But, uh, and, uh, well, great.

Yes, uh, I know, uh, we'll certainly be sharing, uh, thoughts on Mobile World Congress and, uh, beyond that, uh, thank you everyone for joining the 5G Factor. Again, you can bookmark us on the future and group, uh, website as well as we can be, uh, viewed on Tech strong, uh, tv. And, uh, we certainly, uh, again, appreciate, uh, taking the time to listen to our thoughts.

And with that, everybody have a great AI and test time scaling as let alone 5G day. Again, thank you all. Hey, everyone.

We're all trying to figure out how this AI thing is gonna work in the enterprise. At the same time, we're faced with all sorts of new applications, new technologies, AI code generation, and yet another model coming to, uh, the, the world from China. You're watching Textron Gang.

Hey, everyone. We're coming to you live from San Jose, once again here for the Techron Gang. I'm Steven FoST, organizer of AI Field Day and the Tech Field Day events for the Futurum Group.

And also, you might recognize me from Tuesday episodes of Textron Gang. I am not Alan Shimmel, however, but I'm pretending to be be him for these, uh, couple of episodes that we're recording live here. We are joined in San Jose by a great panel of independent technical folks who are, uh, participating, asking, uh, questions, discussing, uh, with our AI companies during AI Field Day.

You'll catch all of those things live streaming on Techstrong tv. But let's meet who we've invited to be on the panel with us today. So first up, uh, we've got, uh, a special guest.

Uh, once again, we've invited her back. Uh, Gina Rosenthal is one of my favorite, uh, tech thought leaders. I'm sorry to say that to you, Gina, but oh Lord, she is absolutely a tech thought leader, uh, joining us here on The Gang for the second day in a row.

Woo. Good morning. How are you doing this morning?

Well, it's okay. Uh, bad news outta DC Boy, that, uh, oh my gosh, that was some crazy, crazy stuff. That's, Yeah, that's just hard to even think about.

So, Uh, also, uh, we've got another great, uh, gang member here, uh, John Willis. Um, great. We'll take it.

We'll take it right. Take 'em when they come, right? Yeah.

Yeah. Right. Hey, yeah.

Great to be here with, um, you know, uh, the Tech Field Day has been great to be involved and doing, uh, tech Gang. Yep. Here live, it's great, great stuff.

And of course, uh, familiar face. Mitch Ashley. Always good to be here.

Good to be bumping elbows and talking to AI. And, you know, out here in the real world, I dunno if if Silicon Valley's the real world Yeah. Real, yeah, yeah, yeah, yeah.

Sort of maybe kind of thing. Good, good to see everybody here on the gang and welcome Gina. Good to have you, and thank you.

Yeah. Yeah. Be back with you too, Steve.

So, we, uh, yesterday, uh, had, uh, tech field day presentations by a number of different companies. Um, the, the, the big ones, uh, that we saw that were sort of, uh, independent third party product presentations were, uh, from VMware and from member. And it was interesting during those sessions that we had, uh, both, both companies, at some point, some of the delegates internally were saying, Hey, these guys are trying to be the VMware of ai.

Now, it's no surprise that VMware or Broadcom more specifically, it's no surprise that Broadcom's VMware team would want to be the VMware of AI because they wanna be the v VMware of everything. I mean, remember, they wanted to be the VMware of storage. They wanted to be the VMware of networking.

They wanted to be the VMware of cloud. And so that makes sense. MERG was an interesting one because, you know, previously I had seen, uh, their, their technology, uh, we had witnessed what they're doing.

Uh, they have some very cool tech, but I didn't understand the scope of their ambitions until they presented yesterday. And it became very clear that they also are trying to become the VMware of ai. In other words, they're trying to build a system that is scalable, that allows sharing of resources seamlessly, that maximizes your hardware investment.

And the big thing that brought that home to me was in the first few minutes of their presentation, when they showed that chart showing that only, uh, you know, that, that, that the majority, uh, a third of enterprise AI installations are only used using about 15% of the compute resources they have available. That was pretty mind bending. And it shows the opportunity for somebody to be the VMware of ai.

'cause you know, Gina, we were there at the beginning when VMware first started. Mm-hmm. The reason VMware took off was because essentially the same stat.

Most enterprises were using very little time of their resources, and there was an opportunity to do convergence. So, uh, talk to us a little bit about what you see as an analogy. Well, first of all, I don't think VMware's trying to be the VMware of ai.

I think they are the VMware of ai. So they were, they've been working on, and this particularly when we're talking about merge, is, um, the GPUs, you know, you spend so much money on A GPU, you need it to be able to have the compute go fast enough to run all of these, um, jobs that, uh, building an AI platform will require. Um, so VMware's been working on that.

You can, you can virtualize a VM or get to a, I mean A GPU or get to a GPU three different ways. And that's been true for, since I was at VMware, so seven or eight years ago. So they have done it, but that's what they're looking at is how do we virtualize have a virtualized machine?

How do I get a slice of A GPU? And they also talked about wanting to make sure that, uh, the GPUs were being utilized across an organization, and not hoarded, but shared by the vSphere Management system, which all of that does make sense. Um, and I, I think member is doing something very similar.

They're looking at how do you provide, it was really cool. How do you provide fractional access to A GPU and how can I put that in a management plane? So if I have three and you have six and you have 10, we have put the whole lot of those that the company owns into a, a platform that we can share the GPUs across.

And then if it gets too busy on one, we can move 'em off pretty transparently, which is an awful lot like vMotion of GPUs. So, um, I, I think that it comes back to we we're looking at how do I make, instead of, uh, the compute on a CPU on a server shareable, how do I make those GPU shareables, those GPU shareable, because they're so expensive to get so critical to ai, uh, workloads, Think there's a spectrum of, of what they work on. And to me, VMware is how do you extend VMware into the world of ai, AI applications and AI hardware, whereas Verge was, you know, in lieu of creating the next A MC show called CGPU Hoarders, let's figure out a way to, to do GPU sharing and, and their approach.

They started with talking about the developer access through the IDE and being able to schedule that as part of your dev workflow. Um, and they were, and then they were adding all the things of snapshotting and, and restarting across different hardware, uh, which kinda led it to look like a bit of, of VMware. But my thought was, I don't know how member becomes the VMware of, of VMware of Gen ai.

'cause they're already gonna be it. That's, that's my, so they've gotta bifurcate somebody. Well, My point is, it's an apples and oranges conversation because GPU as a service and in infras service is a piece, and it might be the most interesting piece, giving all the constraints and things we've learned and we know about.

But the real problem is how do you supply AI in all of its spectrum, from containers, from virtualization mm-hmm. From, you know, things like Harbor and let's not forget networking you NSX, you know, uh, or, you know, I go down the list, right? Go governance monitoring.

So when you talk about the, the Apple, which is Verge, which is a really interesting solution for a very interesting and hard problem. But there, when you start comparison Broadcom versus emh, you, you know, I I I pointed out that like the, I think you could use the analogy of do you run OpenShift or do you just piecemeal Kafka? Do you run OpenShift?

You can get Kafka, you get policy, you get monitoring, you get all this stuff. And I'm, I'm, I, you know, I go back and forth depending on the client, or do I do Kafka upstream? Do I do, you know, do I do, uh, you know, Kubernetes upstream?

Do I put in all the monitoring myself? You know, and I think Broadcom's gonna have a more compelling argument because the technical debt of doing all that yourself becomes a problem. Well, and, but I don't think Broadcom is gonna attract new customers become, because they become the AI of VMware.

I think it's the path for VMware customers. Exactly. Yeah.

Yeah. And it's an easier sell to walk in and say, Hey, do you really want to have a, a Kubernetes for this, a Kubernetes for that, a Kubernetes for that? Or do you want to have a control plane?

Mm-hmm. That we kind of invented. We already got it.

Right. And The thing, the thing about that is that, that's what you have to buy. So if I'm a, if I'm a broad current Broadcom customer, I know that I can only buy the whole kit and caboodle.

Right? And it's expensive. So yeah, it, it might be the best in class and have everything all there, but you're still gonna have to pay for that.

If, even if all I want is to have the ability to, uh, democratize the GPUs in an organization. For me, what I thought was interesting, I kept thinking sounds, when they were talking about, um, when member was talking about sharing the GPUs, I kept thinking about, this just sounds like SRDF to me back from the EMC days. And then having a conversation with the CAOI find out, oh, you're a storage guy, so you're approaching this as, how do you know we've got this data sitting over here, we gotta get it in really fast over here.

I've got this equipment that's slowing me down because I can't get to all the equipment. How do we solve that problem? So I think that's an interesting problem.

And I think if, you know, you're talking about they've got that they're sitting on top of a layer of Kubernetes. So if I've infrastructure as co infrastructure is coded, all the things that I have, I should be able to build this pipelines that get, I, that's another thing, right? When you talk about, like, the things they don't, didn't show with, uh, Broadcom yesterday is when I asked like, what script?

Well, there's years of Terraform in there. Yeah. Like, that's not gonna show.

I mean, you know, they're, they're based on the questions Verge, again, I'm rooting for 'em. They were very immature about what it was gonna take to manage Kubernetes. And, and that's gonna include a lot of stuff that historically has run at scale.

The only thing I, I mentioned too, I, I talked to the, the CEO last night too, and I, you know, I'm gonna give you unsighted, you know, three margaritas. So I'm like, yeah, like I think you need to go open source. You know, you need to become the lang chain.

If this is gonna something you're gonna solve, you need to get in the middle of everything. 'cause you're gonna have such an uphill bo battle dealing with the Broadcom messaging and whoever else is gonna come out with this, this sort of like, full solution. You know, red Hat obviously is going to have some, they do it in Instruct Lab and stuff like that.

So, you know, in order for them to get, like, become the lang chain of, of this new stack, you know, so I, I don't know if you, You know, well, it's like sitting between, you don't need to be VMware. You already have that. Mm-hmm.

You, you're running on Kubernetes, I'm not sure they know what they've got there yet. Right. Right.

All the capabilities that, that, uh, has as well. So what they can do, but they want to be the stack on top of it. Right?

There's a lot of companies pushing for the, how do I get AI into production? There's the neural magics that, uh, red Hat bought, right. Very much down that lane.

And these guys could be, uh, on that same path. And I think there are other, I mean, it's, I think it's gonna be a very competitive, because Kubernetes just step in and solve all of it, right? No, it's just a workload.

It's control plane. It is just a control plane. Exactly.

And, and it may not be a great control plane for ai. You know, that was one of the things that came up in the discussion was that, uh, you know, many of the aspects of Kubernetes that makes it compelling for, uh, webscale modern applications makes it not all that compelling to run, uh, machine learning training and inferencing and so on. Especially The training.

Yeah. Yeah. And so it's gonna be interesting to see where that goes.

But I think that the reason for me, I, I think I'm basically just exposing my tech nerd, hardware nerd background, because, you know, when I look at what they're doing with the fractional GPUs and GPU sharing and splitting and optimizing infrastructure utilization, it does, it reminds me so much of the messages that we got from classic VMware way back in the day, really, you've got this incredible investment in hardware. It's being underutilized, you've gotta utilize it. But I think you're right.

I am, I'm gonna say that, you know, number one, we have the VMware of ai, it's called VMware. Yeah. Uh, number two, uh, member is a very different solution.

And, you know, more of a, more of a, a technical solution that may end up finding its way in a completely different way than, you know, becoming some sort of, uh, you know, hypervisor for training to your, To your point, I mean, they're starting at, obviously at different places, but they're talking in different audiences, right? You're VMware's talking to the running very large infrastructure. Mm-hmm.

So they're solving the AI of VMware from that perspective. Me is gonna be talking about AI development and model training, AI engineering teams trying to do work to get things into production. They're gonna solve different problems.

I, than I think VMware will ever get to. And I Don't know if I fully agree with you. I know we talked about it earlier, because if you, they did have, they did focus on the assists admin, the DevOps, right?

But, but that was to create the developer experience, right? So I think there is a path of like, development experience, and they did talk about, you know, sort of GPU management and there was a difference between how they did that. So, and, and the other thing, I mean, again, I'm, I'm not like well versed in, you know, CUDA or Nvidia APIs all, but like, it sounded like Verge is really just using most of what, um, NVIDIA provides.

Mm-hmm. Right? There's no, like se you have to work with their management tools, their APIs, and, and so, like, to me that just says Broadcom will be able to catch them as quick, you know, if it's all exposed and they're just using, what's the IP that comes out in nvidia?

I'm not sure. I think there might be a race problem. That's why I think they're gonna have to find what it is they're gonna pivot to.

That was the, one of the que they really have to be clear about what exactly are they doing. Uh, and, and It's a very interesting product marketing problem from a messaging perspective, because both companies, the tools are, uh, are used by operations people. And operations people don't really, you know, necessarily, none of us really have the vocabulary and the understanding for how an AI workload works.

It's, it's just a workload. So you should be able to manage it and make every, all of the hardware perform it and do everything the developers want to do. And that's the, the goal of now a platform engineer, which is assist admin in my view, but, you know, help me, I'm wrong.

But to get the platform engineers to a place where they have a self-service platform for the developers. The developers and that team, the product owners are running that show for how we want to share our GPUs or how we want to do things. So you may have this IT person that's connected to the IT department that's specifically trying to figure out how do I get them to stop fighting over the GPUs?

How do I make this so I can tell when the server, the GPU is attached to is gonna go down and I can provide all the security and data protection and fault tolerance and everything that I need to provide to keep a business running. But they've got a really weird messaging problem where the people that are gonna be using it and need to understand how the, the developers work, so they, they can serve that team, have no say in the buying at all. And that's, that's a really hard problem.

So you have to have, that's why they both led with this. We know you got a bunch of GPUs sitting out there underutilized. That's a lot of money you're wasting, let's fix it.

Because that's a problem everybody can underst. And, and I'll tell you this. So I'm glad you worked that out because one, I, I worked, I was early in a docker, right?

And Pivotal killed us because you know who they sold to, the infrastructure people. Mm-hmm. Right?

And the infrastructure people who were by, that's gonna be Broadcoms, they're walking in, we were saying, well, should they be more dev? The most dangerous thing you can do is take the Docker out where you tried to sell you. You believed that if you could get billions of developers using your tool, you'd win.

And it, that, that theory did not play out. I mean, Docker won, but Docker as a corporation lost. And, and, and it really was because the competitors of Docker were focusing, you know, pivotal.

Um, you know, then, um, you know, Mesosphere, like they, they were coming in at the infrastructure level. And Broadcom's walking into your point, they already have the infrastructure and you have to win the infrastructure. They have it.

Yeah. And they already own It. And Broadcom, that's the thing.

I think that's important here. Broadcom, when bought VMware, uh, Broadcom already was a leader in infrastructure. Now they're even more a leader.

Yeah. Yeah. I mean, Broadcom with VMware is even stronger than VMware ever was in that market.

And you're right. I think that that's where a lot of the money is. Yeah.

That's where a lot of the buyers are. And, um, you know, and, and so I, again, I think that shows that these are not like competitive products. These are two completely different products, even though the messaging is similar.

Well, we do have to move on, uh, to our next segment. And in our next segment, we're gonna talk more about developers, uh, keep watching Discover Techron Group, the epicenter of tech innovation. We are your go-to for reaching IT leaders and practitioners worldwide.

Our secret impactful content that sparks awareness, engagement, and top quality leads with us. You'll access editorial websites, streaming videos, virtual events, custom content analyst research, and more. Join our satisfied clients.

Let's revolutionize your tech journey. Contact us today and tell your story to the world in the most powerful way with Textron Group. Welcome back to the Textron Gang.

So we are going to turn the page here from, uh, infrastructure to developers, and we're gonna talk about one of the primary applications of ai, which is coding. Now, I think that there's this thought that somehow, uh, chatbot can just spit out a functional application with no real input, and we don't need all those developers anymore. But I'm not sure that's really the case.

So John, uh, talk to us a little bit about the landscape of AI coding tools. Yeah, no, I mean, you know, there was the Che GBT moment, then there was the co-pilot moment, which sort of were really close to each other. And they, they had similar effect on different groups, like the developers.

Immediately, a lot of people are like, this is nonsense. This is terrible. You can't do it.

It'll generate terrible code. And, and that was all baloney because I, you know, I talked to banks that have 18,000 Java developers, and I'm like, what percentage of them actually create great code? You know?

And like, so let, let's get off of this, that the AI agents create bad code, right? Like, it actually creates better code than most of the coders in your organization. So then the copilot was interesting, but then what came out, if you've heard Cursor, cursor sort of blew everybody's mind because they, somebody got their 9-year-old kid to write an application with it, which again, was more marketing and nonsense, but it sort of changed everything about copilots.

Just like gimme Code Cursor was like intelligence in the IDE. And right now there's like three or four products that are really interesting, cursor ada, but it, you know, and now it's getting confusion. Like, what do we, you know, what do I use John, right?

And I, I've been working on this like blog article that I'm trying to write. Like, the truth is you're gonna use more than one, and it's gonna be based on the tasks that you do. Like if I'm just wanted to prototype an idea, I might just do that in code.

If I'm gonna refactor code, I'm going to use probably Cursor or something called ADA or client, right? So, and you're probably gonna use, you're gonna come up with your sort of best two or three. And then you start understanding like, cost.

And then I read something interesting that like 80% of whatever coding tool you use is really comes down to the model you use. 20% is the interface, right? So now you gotta really think about the coding tool and the model where 80% of what you're going to accomplish is based on the model you use and some model combination.

So there's some really interesting, I think as cut as executives and team leads try to decide, should we just pick one product and everybody has to use it? Should we allow them to have freedom? I think, you know, and then there's the sort of scale of the coder.

There's junior coders. Like for me, I'm not an industrial coder that codes every day, eight hours. I like Eighter, friends of mine who are industrial coders, like cursor it, it appear, it appeals to them better.

So the IDE itself, the tool, um, so any, I, I think there's just like token costs, use cases multimodal. And one last thing I'll say is context is interesting. Some of the tools actually allow you to pull the full context.

You know, what happens when you go in a copilot? The context is just code. And that's not really how a software engineer builds an application.

No. They have documentation, they have Terraform scripts, they have all these non functionals. And if the AI tool is not aware of those things, your sort of, uh, collaboration with the tool is limited.

So I think what we're finding is some of the, these tools that really allow to expand the context reads in a repo and understands not just, you know, run into how it was built, what are dependencies, you know, what, you know, what's the documentation, even Jira tickets, right? That kind of stuff. So some of these ones, and I'll just, you know, ADA, that ADA seems to do a really good job.

And there's a new one called Windsurfer. And then as you pointed out earlier, in all that craziness, Monday of, uh, deep seek, um, bike dance, put out something called Trey. And I haven't, I'm for obvious reasons, I haven't tried it yet.

The same reason I didn't use the API over at Keepsake. But Well, I, I think you did a really good job of kinda laying out the, sort of the emergence of all the different tools, right? That are built on what kinds of models.

Um, and a couple things that you eliminated. One is it's context as in what's the whole code bank? What are the other things that developers do, right?

What, what does that whole thought all mind process of, of managing and delivering software, not writing a piece of code. It's like the equivalent of what model generates the best code. It's like, which model generates the best TV script versus a blog post, right?

But you need really the full context of it. And a lot of what drives drives it is graph technology that interconnects, here's your, here's your Jira tickets, here's your open items that you're supposed to, here's the, the technical debt backlog. Here's five ways that we do have implemented this particular coding pattern in our software.

What's the best application of it for this situation? That's what a developer looks at, right? Yeah.

What do I know about how we do? So, so we're, we're still at the experimentation stage, which is a lot of the choose the best tool for whatever. Yeah.

And still focused on the point of generating code. I think the next step beyond that is really taking it to it. What's a real workflow look like?

And it isn't black or white, it isn't like cursive versus this versus code. It's like, is it a quick fix? Is it a refactoring, which is a big job.

Like the different tools will have better workflow for the type of thing you do. 'cause the developer might be somebody who's just fixing to do commits and development might be refactoring like, you know, 40,000 lines of code. Those are different type of things that require the tools react differently for those.

So to your point, my best advice is do the bakeoffs. There's some great video bakeoffs out there. Literally try to just take an application, take a source code base that you know really well, and try it in different scenarios.

Quick fix, refactoring, uh, from scratch, you know, new componentry, new features. And then you'll get a feel for which of these tools actually are gonna work Kind of beat trip ticket engineering approach to it, right? Duh.

Yeah. Like who Go figure. Yeah, very much so engineering.

Yeah. Well, and there's others as tab nines, there's also companies who are doing a lot of work around analyzing code bases, like for app modernization, whether it be mainframes. Yeah.

What job or whatever it is. Just telling you what this code does. What are the, like what, what it's, what is it solving, what Function?

Well, the good news is a lot of those have all that built in now. Yeah. And that's combining those things.

Exactly. Right. So it's the, the point of the article that we kinda surfaced that started this conversation is develop, well, how much time do they spend writing code?

But also, how much time are you gonna spend understanding and fixing code that's generated for you? Right. Or just, you know, you know, the developer days of I'll just do it myself.

It's faster. Right? Well, when that goes out past that productivity hub, no, I'll just take what they gave me and I'll fix it when I need to fix it.

Right? I'll use the tool to help me understand what it's doing. You know, that sounds terrifying to an person because work worked in dev is a thing, right?

Hey, it would've worked in, Would've worked in production. Yeah. But I mean, No.

And, and, and That doesn't change, right? No, It doesn't sound like it's gonna Change at all. We Think It just happens.

Stack overflow. Cut, cut and copy and paste to now, you know? So, I mean, because like, give Me Yeah.

Like I could get wrong. Yeah. And, and I think that one more thing that I'll throw in here, and I think probably Gina's thinking about this too, uh, because I know, I know what your, uh, your concerns are about a lot of these things is the privacy aspect.

So when you're talking about, you know, oh yeah, you know, maybe you put your code base in this one and try it out. Put your code base in this other one and try it out. I'm like, wait, whoa, you, you're gonna do what with your source code, you're gonna put it into, you know, your proprietary enterprise source code and all of your commits and all of your bug reports and, and, and everything.

You're gonna put that into an AI engine that is, uh, you know, and so cursor, for example, uh, claims you know that they're soc two compliant and that they keep everything local. I like that. Um, not endorsing it by any means, but I'm just saying, you know, I'm glad that they surfaced that right away by the, but, but that's my concern is, you know, you put your code into these things.

How do you know where your code is going? Yeah. You don't.

Well, but when you're enabled, uh, Google Gemini on your Google Workspace account has access to everything in a lot of that's happening with it. It's happening all over the place. It doesn't, it's a legitimate concern.

Doesn't, But the thing about it is people don't know. So this is going back to like, the ops people don't understand the architecture. So we're having a hard time figuring out how do you, how do you manage everything?

But I think for all of us, like what if you, if you put something into Google or Microsoft and you don't have a paid account, at least with a paid account, you've got at least the, I dunno if it's a facade, but you got the, they tell you that they're not gonna do anything with your, your per personal data. If you go to some of these other new ones that are popping up, all bets are off. And I think people need to educate themselves about that.

It's so important. Gene, I don't know if this will, this will help, but I can imagine a day not very far down the road that because we understand your code base, AI does, maybe it generates some of the code, some of the not and ops person can say, why is Kubernetes doing this? Absolutely.

And it isn't gonna say, well, Kubernetes has these three features and here's how they work. It's gonna say, here's what your code, our code is doing, and here's three things you might do to That. And I think that, can you imagine that it, it worked in dev, it's not working in ops because of all this, all the other things we have to do in ops to lock everything down.

And I can just ask a question and it tells me. Yeah, that's amazing. And, and, and again, the models itself, you know, like, I mean, and I, I wouldn't trust like NSA code to this, but like OpenAI, they don't, they state that we not trained on your data.

Right. Like I still think there's dragons out there of putting anything in context, but at least there's model providers that clearly state we do not train. They know it's an issue.

Yeah. But you have to pay for it. And in my experience working with devs, they, I mean, I, I used to work one place that I worked, um, we had, it was astrophysicists, postdocs, and they had to, um, get on our network and we'd get a ticket if they tried to plug in, and we'd go and help 'em out, help them, you know, fill out all the, protect the forms they needed to have.

One time I went to look, and the dude, I swear to goodness, had like an all cardboard box with a towel in it that like, you put lost kittens in, and he had breadboarded his Linux machine together. And I was like, this was way before like cell phone. So I don't have a picture, but I just turned around and went to get my boss.

Like, I'm not even Like an academic Research environment To open Things up. The good news is, a lot of the large corporations that I've been working with are really setting standards on things like they have to, what, here's the only copilot you can use. And it's already, you know, so, and you know, again, do you, I'm sorry, I'm laughing at that.

Just because any, I mean, setting corporate standards for anything is not always that effective, especially when it's as easy as going to a website. I, I agree. But they have to do that.

Otherwise, it's just a free for all. You know what show? Oh, I, I'm not saying they shouldn't, I'm just saying that it's not gonna work.

Well, it works in the sense that as we get, you know, there's another whole subject, but, but the whole sort of auditing and, and I think what's going on in, you know, what we wrote about Investments Unlimited, that mm-hmm. What's going on there about making sort of DevOps auditing, if you will, para phrase. Um, it, it's putting more light on the things that you're not doing as an organization.

We're informing internal auditors way better now than we were pre five years ago. And that puts a flashlight on use. 'cause you have to document the chain, the supply chain of what you did.

You know, the, how did you get from your code to deployment to all that stuff. If you're sort of creating attestations around that Fantastic book, by the way, Download the, that allows flashlight for the organization to start finding those Shadowy. We have to talk to the individual, everybody, because right now there's so much hype around ais and models and do this and do that.

And it's exciting and it's, oh my God, I wanna go try it. And I think there needs to be a little bit of education of like, I know you wanna try it, but really, really think about do what will happen if you connect to this API. Mm-hmm.

What will happen if you do anything with data you really care about or that your boss might really care about. Yep. Yeah.

I, I, The hype is It somebody said, uh, I think it was John, uh, that this was the greatest data exfiltration Oh. Monday, um, in, in history. Because by dance or, uh, not by dance, because deep seek got so much attention that, um, you know, probably literally millions of people register, dumped all sorts of sensitive information into it immediately.

And that was that. Yeah. TikTok went, why didn't we think of that?

That's a good one. So we gotta move on. Thank you very much.

Uh, this is a conversation I know that's gonna continue on the Textron Gang. Uh, up next, uh, we're gonna look at, uh, new model, uh, coming in from China called Quinn. Stay tuned.

com is the leading resource for news analysis and education on challenges facing the cybersecurity industry. com covers all aspects of cybersecurity, including data security, DevSecOps, cloud security, application security, network security, security threats, and more. com has the largest selection of security content featuring breaking news, blog posts, podcasts, and more.

com to learn more. com. Home of Security Bloggers Network.

Welcome back to the Textron Gang. We are at AI Field Day, and, uh, if you've noticed, every one of our conversations has been about ai, but that's not all that unusual. I think everything on every episode of Textron Gang for the last couple of months has been ai.

Yeah. Uh, let's talk a little bit about another new, uh, model coming outta China that claims to be even better than deep seek. 5.

Max says that it outperforms deep seeks R one, and that it uses, uh, low amounts of energy and all this kind of stuff. Essentially, it's the same story again from, uh, Alibaba, which is a huge, uh, Chinese cloud company. Uh, who wants to take this to start?

I, I, I'll go just quickly. I mean, Quinn has been around for a little bit, um, so, uh, and there's been versions of it. Um, you know, Alibaba's real, I mean, you know, Alibaba Cloud is real.

It has been real. It's been, you know, there, there have been a number of retail companies that used Alibaba Cloud because you couldn't use Amazon in China. So if you were Starbucks, you literally had to bifurcate your cloud infrastructure between sort of Amazon and Alibaba.

So, Alibaba has been, you know, like sort of vetted to the quote that any company based outta China can be. Um, the Quinn models are, are interesting. If you follow Ruben Cohen, I'll try to get some show notes for y'all.

He's done investigations going back a couple of months now on the, the, all the China models, including Deep Sea, are very restrictive. Like, so they create inference guardrails on things like, you can't ask about Tanam and Square you. And the, that, you know, that's just the obvious.

The unobvious is when you put those constraints in an inference model, the algorithm type models that you do, it actually creates more bail for Jailbreaks. And Ruben was able to show some really good examples of how you can use inference jailbreaks to get it to ask questions that the model itself is designed not to answer, you know? So, um, so again, I think this is a, this will be a problem that will bubble up on a lot of the, the models that come outta China.

The fact that they create restrictions in the sort of algorithms themselves or in the inference, actually makes them more vulnerable to Jailbreaks. I wonder if it, if it sort of sets the race, the space race to get to general AI a bit to the side, right? We're all focused on kind what the next model is.

That's who's gonna get to the singularity first. And I was making this point the other day. I think it was on Monday or Tuesdays show, talking about, yeah.

But there's also good enough ai, not everybody needs, you know, the singularity to solve every problem. NGI and I are full slacks. Yeah.

They can, you know, the the, it it takes us, it's a distraction. It's a distraction from absolutely what do you need for ai? Let's stop worrying about whether it's gonna take over humankind or Terminate.

And it's, and it's not a Scientist. Let's be engineers. Yeah, yeah.

With what we, we got It's not a single point. Isn't, isn't general AI gonna be Well, like what? Q like Q General, I'm steal that.

My general AI is smarter than your, anyway. I think that's sort of a Yes. That's part of the race.

The money's made. 'cause nobody's gonna be able to afford to run it, right? I mean, that's gonna take the, the highest compute of anything.

Every, what's interesting is the application of AI and how accessible it's to people. And I think that's why TikTok being jealous of, of, uh, a deep seek. I called it deep fake the other day by accident.

Oops. Uh, I guess Freudian in some ways, But tiktoks got its own stuff going on. It's the cult.

They're after the culture. They're not after they, They are, but they're alsot quite, they're also bite dance though, by the way. It's all coming from Byan.

So, yeah. So, So for me, I think if I look at it from a product marketing, marketing perspective, it's not an accident that these two models were, um, released one right after the other. I don't believe, especially a few days after the us um, government had a huge announcement about, uh, investing Stargate tons of money into Stargate.

And this is what we're gonna do as an, as our country to, to, you know, get up on AI and, and get all of this done in, in this side of the world. Um, it's also, they announced yesterday was Chinese New Year. So that's a great big celebration day.

I think that's a thing too. Um, and I, I kind of find it, it, it's all around the hype, right? Which is what marketing does.

We drive the hype to get everything excited, everybody excited about it. But the, the market is not the same in China as it is in the West. It is controlled by the government.

Everything that happens is orchestrated by the government. So, and, and allowed or not allowed by the government. Mm-hmm.

Mm-hmm. So, very interesting that, you know, we have to start, we have to look at it. China is a, a foreign adversary to America.

And we have to look at it from that perspective, because we are in America. We're not in China, and we are a global universe. And we do want to cooperate and, and be colleagues with all of our people, all of our friends, um, from other countries at the same time.

They're not playing by, they're not playing the same playbook that our company's played by. And the playbook has changed too, right? Just as an author and, and like, you know, um, five years ago, if I was offered to speak in China, I'd go like that.

'cause you sell crazy amount of books. When you, if you're an author in China, I won't go near that place right now if you read about people being detained. So the, the climate of even the global economy within the last three or four years has changed these as a person who speaks, or you, I know we all speak around the world, right?

Right. Like, I wouldn't go, I would, there's no amount of money I would go into China, right? Well, And I'm not, and I'm, and I'm not saying that as like a down thing.

I'm just saying, you know, let's look at this from, if I'm looking from a marketing perspective, what are they trying to do? What are hype are they trying to achieve? Because we know that there are definitely some constraints as far as privacy goes with those models and sharing information.

But, so they wanna get it hyped up, get everybody excited about the next model. And they have done, let's be honest, some really cool engineering work. That part's amazing, but why is it so hyped?

Why is it so, why did they pick the timing? And what does the, is what does the Chinese government have to, to win from that? And, and also, I think we should point out too, as as was said at the beginning here, uh, deep seek didn't come out of the blue.

Uh, this was, you know, just the latest iteration from a company that it, you, you know, you would've seen this coming if you had been watching closely. Uh, the same with Quinn, uh, you know, as we just heard, you know, there's been other Quinn models. Uh, this is just the newest one.

Okay. Um, it seems pretty good. And, and as we talked about yesterday, uh, when talking about deep seek as well, uh, in many cases, the US uh, restrictions and controls on, uh, on AI components, forced them to build these models in a more efficient manner.

So just like deep seek, you know, the Quinn is a mixture of experts model with, uh, that uses a lot of small models kind of teaming up together. Um, it makes a lot of sense. You know, this is the, this is the microcomputer versus the mainframe in, in AI sense.

And, and so it's not really a surprise that this is, that this thing came, and it won't be a surprise next tomorrow or next week when another model comes out claiming to be, uh, deep seek or chat GPT or, or whichever the benchmark is. But ultimately what we're talking about here is basically who's got the biggest, baddest, fastest, strongest, when what we really should be talking about, I think, is, uh, the application side of things, which is what we were talking about earlier on this discussion. Well, elevated a level you mentioned Mar the marketing of it, it, it's, it's geopolitical Yes.

Marketing of it, right? Because, because China's the way it is, then it creates all kinds of conspiracy theory or reality, whichever you choose to believe is, is this the Star Wars moment for China of setting this big thing that now we have to go all go chase them after, kinda like Reagan did with Star Wars back in the eighties, right? Or is this something legitimate?

I think, by the way, self, self shameless plug. Either today or yesterday, um, rum research came out with a paper, our kind of position thinking about it, different parts of Oh, cool. The deep seek announcement Daniel Newman contributed than I did as a bunch of other analysts.

com. Well, we do have to wrap up now. Uh, thank you.

Sorry for Joining us. I your month. Wait, You Got one?

I have one point. Oh, Okay. Hit hit me, John, Then.

There's no doubt China is an adversarial nation. But, but yet Monday and Tuesday, we heard a lot of, um, well, China, you know, oh, there's America and China. Well, you know, the stuff that came outta France, Mistral.

Yeah. So there wasn't, so, and on the geo, geo computer science spectrum, this is just d different countries that are just showing that they have ip, they have knowledge. But I can't exclude the fact that like, I'd be more trustworthy of Mistral from France than I would from Quinn.

When, so when did Myst come out? That was a year or two years ago. Okay.

That's what I'm saying. Like, I think it's very interesting marketing. No, good timing.

Oh, I totally timing. I totally agree with you. Timing's everything.

It's interesting. Yeah, I totally agree with the timing point. Anyway, anyway, I was saying in awe because I was sad.

We were, it was over. Well then we will, we will have to, I'll, I'll, uh, put the screws on Alan to invite you back and, uh, I get to meet Alan. That'd be Awesome.

That would be amazing. Um, so thank you very much for watching. Uh, tune in, uh, uh, Textron tv for all of the AI Field Day sessions.

Uh, we've got some great presentations, discussions. We're gonna do a round table discussion. Uh, we've got the, uh, weekly Tech Field day podcast streaming on Textron tv every Tuesday.

Uh, and of course, uh, check out the Textron Gang every day, every weekday. Uh, you'll find, um, at least three of us, and I'm gonna guess four of us, uh, back on the Textron Gang in the, pretty soon in the coming weeks. And, uh, we'll see you there.

So thanks for watching and Atlan, um, uh, come on back. You're our only hope to rescue us from, uh, from my hosting. Uh, take care.

This is Textron tv. Hey, everyone, we're back. Techron tv.

We've got another CEO to talk to. Let me introduce you all to Tom Fin. Fin link.

ai. ai, but hopefully after today you'll know who they are and what they do. Let's welcome Tom here, though, to Textron tv.

Hey, Tom, how are you? Thanks for joining us. I'm doing pretty well.

Yeah, how are you? Very well, thank you. So Tom, we're gonna talk a lot about Conifers, but before I do, let's talk about you, you know, you're the CEO here.

Give us kind of, how did you wind up being the CEO at Conifers? Oh, yeah, that's a very interesting story. So, you know, started my career, uh, back there in Tel Aviv as part of the intelligence community over there, doing a lot of very interesting, exciting things we can talk about.

Um, then after my military service, I joined a little company called VMware. Um, back in the days, uh, when we did public clouds, private cloud, and a lot of interesting things in the data in the data center, um, I had a pleasure to join the Israeli office back then in Tel Aviv and work on the European market. Uh, but as our offerings to ka um, I was asked to move to California, to Palo Alto, to headquarters, um, and lead the, um, portion of a product team of the cloud management division.

Uh, so this is where I first exposed to data science, and we did monitoring and data science, uh, for the data center about 12 years ago. It was remarkable. It was really interesting.

Uh, we got a couple PhDs helping us, and, you know, we needed to have, you know, mountains of servers in the server X in the data center just to run some simple predictive analytics, um, and machine learning. Um, after that, uh, did a lot of work on move to the startup side of the house and ran product and engineering, uh, for a sustainability startup. When we did optimize, uh, consumptions and generation of power plants in the United States based on predictive analytics, was very, very interesting.

Um, and then I moved to insights, uh, doing, um, Fred Intelligence, uh, great company. Uh, I moved to the dark side, moved from the product and engineering side of the house into the sales side of the house. And I was the Chief customer officer over there, uh, six and a half years.

Great journey 2021. We were sold to Rapid seven on a 350 transaction. Um, and then I joined the detection and response, um, practice at Rapid seven, um, as leading product management over there.

Um, this is where we're exposed to, again, MDR services, the soc, um, and a lot of the different phase that we are doing today at Conifers. Uh, great run over there. Um, one more experience meeting a lot of great people and customers.

Um, and this is what inspired we see the challenges. See how you provide soak services in large scale, uh, and see what's working and not working. Gave me the opportunity to kind of understand what, what we think, what I think what I believe should be the next stage in detection and response and security operations.

Um, and inspired me to kind of, uh, open conifers and start with con first. Excellent. What a journey.

Oh, yeah, what a great journey, man. All right. Only only in the tech world do you hear, you know, these kinds of fantastic, fantastic, uh, career paths.

Um, what made you found or get involved with Conifers? What, what, what was the passion behind that? So, when I was at, at Rapid seven, um, I was implementing a data science project within the Fred Intelligence, um, product line that we had.

And I've seen such a great like results while applying, um, some of those techniques into increasing the efficiency, uh, of our analysts and our ability to do things much faster, in much higher quality. Um, and that inspires me to just, you know, go ahead and see how we can broadly implement that within other departments. And again, obviously being at Rapid seven and being exposed to so many MDR customers and cell customers really helped me to, you know, like I said, better understand, see the challenges, see the needs, see the market, and understand what a great opportunity, um, it would be if we could bring all this like, greatness of data science and the most recent innovations of AI into the stock, uh, and what great problems we can solve.

Um, and that's what brought me into it. Very cool. Alright, let's, let's pivot if we can now with, with Conifers.

Um, you started this company. Look, AI is certainly, uh, on top of everyone's mind, especially this week with what went on with deep seek and stuff. But what, you know, this isn't, this is ai, this is you.

When I look at AI and security, to me, there's two different pieces of it. One is securing against ai, right? The bad guys are going to use AI too, and we need to, to make sure we, we, you know, have a handle on that.

The second, though, is leveraging AI to be more efficient in the security that we do. Conifers seems to me is more of, um, using, leveraging AI to be more efficient, to offer better security, right? Absolutely.

Um, tell us a little bit about that. Yeah, for sure. So I think first we wanna take, I wanna take a step back and just say, AI is great and everybody's talking about ai, but what we're really after is solving hardcore security problems.

So we've focusing on the business impact and the business outcomes on the things that we are doing, and we leverage great technology of making it happen. Like I've hear the buzz around AI kind of, you know, every company as the word AI to every product line that they currently have, but there is a little bit of dissatisfaction in the market with the outcomes that you see, uh, from some of those like products that you see out there. So we are, our main focus is to solve hard security problems through with AI and with data science.

Um, and what we are doing with Conifer is we helping organization to achieve what we call solve excellence and be more efficient. And efficiency has been talked about a lot over the last, you know, 10 years in security operations. But unfortunately with all the different revolutions of the tools that you have seen, that we were able, we were never able to actually see the ROI and get to the point that finally we able to unlock the efficiency gates, but also more effective.

What AI can bring to the table with the right combination with the individuals that you currently have in the organizations is the ability to scale up your coverage, making sure that you catch a lot more early warning signs, and you actually increase your quality and accuracy levels and scaling up in a way that you've never seen before. So what we're doing there with Conifer, it's, we we're taking like holistic look at your stop as it's run today. We help you to recognize where are the best areas of opportunity that you have to be more efficient and effective, and start with a very, very moderated, um, engagement model that help you to have business success and save value.

Um, bring those use cases and onboard them to your specific organization with a great combination between the models that we bring in pre-trained and your institution and organizational knowledge. So we take all of your incidents that you currently have in your SOC today, and we are able to troubleshoot them and investigate them in more depth, in greater scalability, uh, which leads into efficiency, but also greater EFF effectiveness as a result of it as well. Excellent.

Very interesting. Um, you guys recently announced $25 million in funding. Yes.

Yep. Tell us a little about that. Yeah, so, you know, that has been really, really exciting.

Um, you know, that's, that's came up as big trust from our investors. Um, and we're very proud to have, like we've take, like, take this amount of money and make put it into good use. Uh, there are multiple things that we're going to do with this $25 million.

And again, first thing is just our go to market in making sure that we, um, making ourself available to some of the great organizations that exist out there. If it's service providers, if it's, um, enterprise organizations, we were able to go and access them. But on the technology side, we run a lot of interesting things.

We have the cognitive, so this is our proprietary patent pending platform that we created that utilize iGen AI as well as many out of flavors of data science to bring great results when it's come to accuracy, but also cost efficiency, um, into the enterprise. Uh, so keep working on our platform, um, and also kind of accelerating our path with training more models and make the more efficient, uh, and use the, you know, most recent technology in order to bring that, um, bring that great opportunity to be more effective and efficient to a lot of other different organizations out there. Agreed.

Um, you know, I, I wanted to, excuse me, my tongue gets tied there. Specifically, I wanted to mention the cognitive stock platform. This is a trademark, kind of, it's kind of the heart of what you guys are offering right now.

And it, and it's really, you know, it's native AI and it's offered to organizations to solve the critical stock challenges at scale, right? Because, you know, I, I've been in the security world myself 25 plus years, and one of the companies I had helped co-found at one point we really moved into an MSSP type of model, right? We were, or, and I, you know, it's one thing to have a stock for a smaller company, a midsize company, when you start getting a large enterprise or an MSP kind of model where you're managing multiple organizations, you know, every little problem is magnified its tail.

Absolutely. Little, Little problems become big problems at scale. Talk about cognitive sock and how it helps there.

Yeah, so there are few things that we're doing, and I think you hit the nail on the head there. Um, doing ai, it's, you know, great and do AI and small scale is great. The ability to have consistent results in scale.

This is something which is extremely important and consistent results in scale, which are unique to specific organizations. 'cause each organization is different with it risk tolerance, with its assets, with its institutional knowledge. So being able to take that and make sure that the model are being fine tuned to those specific organizations, um, that's really key.

And make sure that you can do it in scale across different organizations. When you talk with a service provider or even, you know, Alan's, uh, kind of enterprise organizations today, there aren't many service providers themselves with the different subsidiaries in the acquisitions that they do. So this is something which we invested a lot in, be in, in depth process of being able to adjust ourselves to the specific technology and institutional knowledge that each one of the tenants or the organizations are using.

We have this continuous learning that we do. And this is another opportunity to say that, you know, humans are not going anywhere. And in order for us to, you know, we discovered a lot of information about the customer from this data, from the interactions with, with the analyst.

Uh, but everything that we ingest into the baseline of each one of our tenant and customers is being audited and validated by human. And why we did that, um, if you would be, if you would lack sometimes AI just to learn things by itself in certain areas without human supervision, especially when we know in some areas that you might have some bad behavior, you might scale bad behavior. And if you start ingesting small bites of bad behaviors into your models, you will scale up this bad behavior.

And that, as you said, a small problem becomes really big overnight. And that's probably would be the end of the AI implementation for that specific organizational service provider. So it's still important to have human in the loop.

It's, it's still important to have human oversight and it's still important to have certain procedures and processes as doing just more information, especially critical information into the baseline of the models, uh, which are relevant to specific customers. Being able to, whatever needs to be controlled, be controlled and audited. 'cause when things goes wrong, first prevents things from going wrong.

And if things go wrong, be able to nail it down, figure out where it is, and eliminate, um, what's going wrong as soon you can. So, super important. I I agreed, agreed.

Con fruit we're almost out of time, but I wanna make sure we get this stuff in Tom. ai, C-O-N-I-F-E-R-S ai. Absolutely.

And is there, like what's the on-ramp? Is there a free trial? Is there Yeah, you know, how, how do people get onto it?

Yeah, so we have, first, we, we do only, well, we sell our software only to customers that see the value in it. So we offer proof of concept and we will love to engage with organization and service provider that wanna come and see the value. Um, it's pretty easy integrating with the existing system platform portals and procedures and processes that you have in house, it doesn't take, uh, long.

And then within 30 days, um, usually, you know, most of our customers able to see significant value. Um, and our unique implementation model allows allow the customer to control how fast they wanna run and how, you know, how what is their readiness to go and adopt more and more AI in their organization. And as we develop that process of being able to, you know, let go for ai, it's also important that we stay on control of that process and making sure that every organization feel comfortable with the speed it takes them to enable ai.

I killed it. Alright, Bob, thanks for being here on Tech Drug TV today with us. I appreciate you explaining all this to the audience.

Best of luck with Conifers, come back and keep us posted as this continues to evolve and, and, uh, grow. Thank you very much, Alan. Appreciate it.

That Alright. Thank you. Tom Fling, CEO Conifer's AI here on Techstrong tv.

We'll take a break. We'll be back in a moment. Hello and welcome to digital CXO.

I'm Amanda Ani, and with me today I have Renee Pbu. He is the director of product at grapht. How are you doing?

Good, Amanda, thank you for having me. Happy to have you on the show. So today's topic is data assurance, but before we get into the topic, can you share a little bit about grapht and what services do you help provide The grapht, uh, network as a service platform?

And what I mean by that is we provide con any to any connectivity, whether it's from your private space to the private domain, the private space to the cloud or, or, or the public domain and the services that you can consume. So it's a consumption based model, so, uh, you don't have to worry about the network itself that's built out for you. So we provide our graph and backbone, uh, as a service.

You consume bandwidth on that, uh, backbone to get you from point A to point B. And we get you to the cloud. We provide you B2B connectivity.

We provide, uh, provide you the data assurance service, uh, all on top of it. And we, uh, provide you the SD-WAN capabilities that, uh, uh, are already prevalent in the industry. So we have a bunch of services that we provide on top of this backbone.

Alright, wonderful. Well, so let's talk about data. And I wanna share, first of all, um, your stance is that data in motion has become one of the biggest, uh, security risk in the enterprise today.

Can you explain a little bit more about that and why you think that? So, data lives in three states. Like, uh, you mentioned data and motion.

The data is addressed, the data is in processing, and then the data is most vulnerable when it's set in motion because for the first time, it moves out of your private domain into the public domain. So data address is sitting still in your domain, in your, within your parameter data. And processing is still within your parameter.

It's when it exits your parameter and into the ether of the internet, that's when it's most vulnerable. And hence, we think that, that, that's the most critical aspect from a security point of view of where data needs the most protection. With the industry going into the AI world today, the data is getting a lot more disaggregated.

It's no longer just your data. You need to share this data with your B2B partners, with LLMs, with GPUs, so on and so forth. So, uh, you are almost extending your parameter beyond your boundary with, uh, places where you're exchanging this data and graphene a data exchange platform.

So with that in mind, we need to really protect this disaggregated data and extend your security posture and your risk appetite beyond just the parameter, uh, into the business domain of things as well. So that, that, that's where our data sharing story, uh, begins. Okay.

So can you share some of the top vulner vulnerabilities and risks associated with data and motion that business leaders should be concerned about? So, data and motion. There, there are, uh, uh, a bunch of things that they need to worry about, right?

One, uh, is it encrypted? Is it, uh, there's a spatial component and, uh, there's a temporal component to every conversation. So contextualizing data is first, uh, the first pillar that's key to protecting your data.

So knowing what the intent of the data is, and then really plotting it on a map, right of, of, of space and time. Am I consuming the data where it's supposed to be consumed? So let's say, Hey, I need to, uh, my servers, uh, are located globally, but if I'm authorized to only access servers in North America within sovereign boundaries, then that's the spatial component of where I should be consuming that data from and not going beyond those sovereign boundaries.

There's a time component of it. If I'm allowed to au uh, access data, I need to access it during workers. If I'm accessing it out of workers, that might be anomalous.

So you need to really contextualize when you are accessing it, where you're accessing it and how you're accessing it. So those are the three pieces of the puzzle that a, any CIO cso, uh, of an enterprise or a service provider would be looking out for in terms of how to protect and what to protect, uh, your data in motion from. Are there any tools or technologies that you recommend that, um, would help safeguard this data in motion Graph?

I, I would recommend graph and the, the data assurance offering itself. Uh, what we've tried, tried to do for the first time is get security to govern routing, extending your security parameter into the network space. Uh, so what I mean by that is data is your most sovereign asset and giving that, uh, the treatment like any sovereign asset, uh, deserves, uh, data should stay within data embassies that extend your sovereign boundaries from point A to point B.

So let's say you, you have GDPR data that's critical to you. Uh, GDPR states that this data needs to stay within boundaries or within trusted entities that extend those boundaries. So keeping that kind of risk posture within your network of where this data travels from Point A to point B, who's the producer, who's the consumer, and those are entities that are GDPR compliant is key.

Uh, and that's the control, uh, graph games to give you, uh, with a click of a button, uh, of where your data moves, how it moves, and who produces it and who consumes it. So would you think that corporate boards play a role in shaping policies for data protections? And, um, if so, how?

Uh, both, uh, uh, corporations, uh, influence, uh, compliance re uh, needs as well as governments, right? And these constantly keep adapting and changing. So, uh, having your enterprise ready and your network ready to adapt to these flex, uh, or these modifications on, uh, data governance and compliance is, is key, right?

It's hard to, uh, rip and replace infrastructure, say, uh, a government changes in a region, the compliance and regulation or encryption regulations change. It's hard to rip cables apart and reroute it around that region if it's not compliant to your, your enterprise. So having a programmable network that can really just add a click of a button reroute that, uh, path around that region is what's key.

Uh, so it, it's a mix of both The regulations are the compliance needs of an enterprise. Example, finance, financial in institutions have PCI, kind of regulations. Governments may have the GDPR and HIPAA kind of regulations.

So, uh, but, uh, businesses have to comply to both. So the, it's a mix of both and giving you that adaptability is key and giving you that control back is key. Wonderful.

Well, um, out of all this, what is, um, what is say one key takeaway that you could leave our audience with today? The key take takeaway is, uh, just protecting your perimeter isn't enough. Having a network that extends your segmentation from your boundaries, from your edge, uh, really getting your business, uh, traffic, the business internet.

Uh, so the internet was inherently built for, uh, open communication, but enterprises need a different kind of internet. That's the business internet and graph aims to provide you that with its, uh, uh, backbone as a service offering, allowing you to program the business internet based on how you are consuming and producing, uh, data and how you govern that data is key. So I, I, I would really encourage folks to focus on how security can really govern routing, uh, in the age of the business internet.

All right. Well, thank you so much for coming on and sharing your insights today. Thanks.

Thanks for having me, Amanda. All right. And thank you to our audience.

Stay tuned. There's more. Hi everybody.

Welcome. We're glad that you've joined us today for another episode of the latest greatest cloud transformation late great cloud transformation. We're talking about really sort of the next generation of how we think about the cloud and the things that we're doing with it.

We're talking about security today, about safeguarding innovation and, uh, strengthening that security. We'll be jumping into app, uh, app security and a lot, and a lot of things here. But, uh, before we get too far down the road, thank you for joining us for this video series.

Uh, the, the last great cloud transformation is sponsored by CloudFlare. We're glad to have them, uh, on board with, with us working on this, uh, helping input with some topics and things like that. And obviously participating on, on our, uh, live editions, which we do on a monthly basis, as well as these recorded episodes.

So thank you for being here with us. My name is Mitch Ashley, I'm VP and practice lead with futurum Group, analyst firm, uh, heading up the analyst area for DevOps, DevSecOps, application development, AppSec, et cetera. So kind of right in, in vain with this, uh, my co-host Alan Shiel is, uh, unattainable, uh, detained, or whatever the word is the phrase is.

And, uh, so I'll be, I'm, I'm hosting both parts of the chair today. Uh, you know, it's a little bit of a coup, but he'll be back next time. We'll see him on our next episode, I'm sure.

So let's get to our conversation, to our topic. Um, let's first start by doing some introductions. I know Chris has been with us on a few episodes here and some different topics.

We've been on other webinars with me and TA talking a lot about application security and, and, uh, cloud Chris Blas, introduce yourself. Oh, I've been in for company my way through the security industry for 30 something years. Uh, I inflicted an early firewall on the markets, and they called border wear, uh, in the early nineties and ran Cisco's firewall business, the turn of the century.

I've been following this inevitability curve and my new series on here on Textron, um, from one spot to another, from firewalls into, uh, sim and network management. From that, you know, the obvious next step is threat intelligence. So I, uh, chaired an IAC for a while, and, uh, supply chain has been my focus the last five or six years, you know, so, you know, software, bill of materials, hardware, bill of materials.

How do we connect all these things, which, and, and currently, so currently I'm, my main role is I'm vice president of Strategy for sebe, which is involved in the BUM space. And I've been, uh, co-chairing several, uh, cisa uh, working groups on SBO m sharing. So we're currently have a group looking at ISACs, um, as SBO M distributors.

How does that know in the middle start taking this information and, and propagating SBO software bill materials? Absolutely. Great.

Thank you Chris. Um, Katherine, Katherine, welcome. Glad to have you on, I think the first time we've had you on the show.

Katherine Newcomb with CloudFlare, please introduce yourself. Yeah, great to be here. I'm excited to talk about application security.

Um, my name's Katherine Newcomb. I live in Denver right now. Um, I've been in cybersecurity for about five years at this point.

Um, and I started in the network firewall space, um, in encryption. And now I'm a product marketing manager for CloudFlare, um, for their application security business, uh, where I focus on their web application firewall product, um, our software supply chain product, as well as our encryption and certificate lifecycle management products. Very nice.

And, and I do like to say full disclosure, Textron, you as a customer of cloud, we do use their services. Enjoyed very much so thank you Catherine, and team for that. Uh, last but not least, another newcomer to our show, Kurt Handel, who's with, uh, Teradata.

Tell us about yourself, Kurt. So I've been working in security probably eight or nine years at this point. Uh, but in the software industry for close to 15 years now, anywhere from development, uh, into business analysis, product management, even, uh, doing a little bit of red teaming myself.

But, uh, I am currently the chief security architect at Teradata. And so I've been focused on architecture mostly for the past six, seven, possibly eight years, and really kind of a generalist. So AppSec is where I spend the least amount of my time where we focus on the architecture, the requirements, threat modeling, um, especially compliance.

We do a lot of the, the major compliance frameworks at Teradata. So we've been pushing that recently. Um, and I'm based in the Pacific Northwest, up in the Seattle area, and happy to be here.

Very nice. All the weather and fires and it's cold and I'm just glad we all made it. Maybe it's 'cause we didn't have to travel anywhere, so, so I hang tight.

I'm glad we're all here. And you know, our, our thoughts go, our hearts go out to the folks dealing with the fires and, and, uh, some weather down south and southeast, et cetera. So, um, let, let's kind of jump in this way.

Um, it, it's a big topic when we talk about sort of the kind of current state of the cloud and where it's moving to. Um, but I don't think it's too much news to everyone that application and app APIs, API first kind of design into applications, you know, it isn't just things that sit at the edge anymore. We think about also the security of the apps and the kind of, uh, software we're creating, the innovation that we're making, um, as maybe as part of the cloud.

'cause sometimes application lives within it, you know, like a, like a provider like CloudFlare or certainly at the edge or at the core as well. Maybe Catherine, if you wanna start us out with, how do you, you're, you're, you're managing, doing product management in this space. How do you look at this, uh, sort of this problem or this space and define it?

Um, so looking at application security, um, when we're talking about this at cloud, we're mostly talking about web application and API security. So if you're an OSI person, layer seven model, um, and you know, when people are accessing these external facing web applications, they're doing it from a ton of different devices and in a ton of different ways. So they're accessing from things like mobile, uh, desktop, laptop, and they're accessing these apps that could be hosted anywhere.

So on-prem, in public clouds, private clouds, hybrids. Um, so as we're securing, we need to think about how can we secure, um, all of these users and the end servers as they're sort of accessing these web apps, right? So how do we make sure that, um, mobile traffic is protected, user data is protected, um, and sensitive data is not, you know, leaving an app.

And then how do we make sure that a web app server itself is protected? Um, so at a very high level, that's about what I think, that's what I think about when it comes to application security. Um, some new things we're thinking about in this space.

Um, I talked about software supply chain. This is increasingly becoming, um, an area of interest as people create more complex apps with more third parties in them. Of course, API first development has also meant we've had to adjust our thinking a little bit around application security as well.

Kurt, how about you as a, as an architect, security architect, you may, maybe you don't get into the innards of applications per se, on application security, but traffic over the, obviously our networks are heavily API driven, um, you know, when you think about the security architecture, where does this fit into your purview? I think it, it fits in really everywhere, right? So we're, we're building these huge applications, sometimes small applications.

I mean, we do all sorts of scale at Teradata. And in my previous roles, I've, I've worked with pretty simple apps all the way to super complex microservices architectures. And so, like Catherine could have said, you have the mobile aspect, you have the server, there's application code literally everywhere, including on the person's device.

And so how do you secure it as best you can, um, within reason, right? Because if it's too secure, it doesn't work. If it's not secure enough, well, you end up in the Wall Street Journal and you're in trouble.

Um, so we, from an architecture standpoint, we really try to focus on all different aspects of it, where the biggest threats lie, um, and then implement controls and use technologies to, to simplify the implementation and streamline it without making it overly complex. And so it's, it's just becoming more difficult given that, um, the, the kind of classic perimeter is gone. Right?

I'm sure you can relate to that, Chris. Oh yeah. Well, it was easy back in the day, right now you had to get on the internet and you needed a firewall.

Get a firewall, right? And I'm thinking as Kurt and Catherine, your comments remind me of these transitions we go through. Like there was the mainframes before our time, but you know, I, I'm old enough to have seen the end of that where all of your capabilities are just to keep one computer running and run terminals and printers and things off that.

And then we get into, or where I came in, where we're starting to build networks, fractally more complicated, just yeah. How do we do that with, when all of our resources were just keeping one computer running, we figured it out, you know, now we're here, we're talking about web APIs, Catherine, you know, the data going in and out and with being stored 30 years ago, you couldn't have that conversation. Mm-hmm.

Now we're saying, alright, what do we do in this case? And it's very complicated. And I think in, and Catherine you mentioned the supply chain.

This is, I think we're filling in the dots. Security has been is not, i i is not new, right? People have been saying, you should know your inventory for a long, long time, and we gotten away with not knowing it.

Now we're starting to fill it in, need to actually know where the software is, where the data is, and we're working through that. So it's exciting times, but it's not different in type than other transitional periods. Certainly is an evolution, right.

Of what we've gone through. And to think about, you know, from the bastion host days early, early on for a pre firewall, um, Well, firewalls used to be a million dollars a year. I think about I got involved, you know, at least as I tell the story, there were a hundred in the world and they typically were seven computers and a team of people.

And my argument at the time was my mom needs one. Yeah. You know, and so we're at this stage where what used to take so much time in here in the API, uh, world has to take less time a lot.

It, it, so let me, let me throw out this hypothesis here. I think it may be pretty obvious, but maybe it isn't, is I think we live in a world, you know, now we we're thinking about things as zero trust, right? Of, of you, you know, anything is susceptible, being compromised and could compromise other things.

How do you pro protect all parts of the network applications, the infrastructure? But we're also living in a world where if so much is determined by what our applications do, not just connecting users to apps, but applications really utilizing the network, being part of the network. It's a dynamic world, right?

It, it isn't a good set of firewall rules and an application firewall, and we're all good, kind of set that up. And it isn't the old days of, if I've got a pizza box in, in my rack for every function that I need, and they're all doing their thing, I'm good. Right?

We need it. It's a much more dynamic environment. So I'm not saying we're reconfiguring our security all the time, but a security has to adapt to, you know, what's happening in the application.

Because we may distribute it to a different part of the edge tomorrow with Kubernetes, or we may, you know, uh, acquire business and suddenly a network has looked much different than it did, you know, three weeks ago. I'm, I'm curious, Kurt, as a practitioner, you know, how do you think about that of, you know, you mentioned microservices and all the things that are being created, you know, in the groups that you're working with. Um, we, we hate for security to be sort of the last thing to be thought of, but you wanna be in the conversation so you can prepare as well as react when you need to react.

I think what you just said is, is really important. You wanna be in the conversation. You don't wanna be doing this retroactively.

And so when you're, when you try to tackle security retroactively, it is infinitely harder to accomplish than if you do it from the beginning. So I have, I do it both ways. I have teams that we work with proactively where they bring us in at the very start and we're building the design with them shoulder to shoulder, drawing the picture in doing security by design or by default as we like to say now.

Or we have legacy applications, which you're doing retroactively and they're quite a bit higher in terms of risk because they've been neglected for so long. Or you find out about something after the fact and it's like, well, how did this get out there? Well, there's shadow IP in a lot of the world.

And so it's, it's hard to, to really kind of put a, a recipe together that successfully achieves it. And then with the, the rapid pace of technology today and how the cloud has just kind of blown this wide open where people can deploy new applications in a hundred different ways faster than ever. How do you keep up?

So you have to implement tooling within reason without doing, without having too much sprawl. You have to have the right personnel partnering with these teams, uh, to ensure that you have coverage and that you, you're really architecting things from the start. Um, and not just kind of using band-aids in bubble gum per se, to, to secure your environment later on.

Catherine, appreciate your thoughts on this because, you know, I remember the days of networks for speeds and feeds and points of presence and connecting A to B and kinda looked like this nice diagram that you stitched together and that was a network and you secured it, now it's overlay on top of overlay and it's changing and mm-hmm. You know, it's, it's multiple pieces that, uh, much more complex to, to secure. How do you, how do you have this conversation with people?

Yeah, definitely. So as you were sort of talking about this, you know, obviously there's a need for responsiveness and customizability and security, but I actually also wanna make the argument for unified policy management in application security. This is something that I've seen actually, for example, um, we have some customers who have protected their SaaS apps, like what is traditionally more of a network firewall or zero trust type use case with the same policy they're using for their web applications.

And by doing this, they're able to do things like make sure that zero day exploits aren't able to exploit their SaaS apps, you know, as well as their, um, web apps. And we see a lot of value out of these unified policy managements. I was talking earlier about, you know, how we have all these apps hosted in different places.

We see a lot of customers, for example, will host, um, you know, an app across multiple clouds for like a resiliency use case. If they're worried about outages, you'll, you'll certainly see that, um, for example. But then how do you have to, you know, actually secure an app that's stored in multiple places?

Do you write different policies for, for wherever those are stored? Um, do you write different policies for APIs versus, you know, traditional apps? Um, so we see a lot of benefit out of like a unified policy for all of those disparate sort of endpoints and all of those disparate, um, locations that they're stored.

Uh, for CloudFlare in particular, how this sort of works out is our WAF is like the backbone, the architectural backbone of the rest of our application, um, security portfolio. And this works out really well because you can do things like have a WAF and an API like positive security model protecting your APIs. Um, so you could do things like detect zero days and volumetric attacks, which are, you know, APIs can also be susceptible to as well as, you know, do the things like bola and, and all those API specific attacks all within sort of one, um, control plane, which we find a lot of people get a lot of value out of because of this really, really disparate environment.

Okay. Chris, I saw a lot of hand waving head nodding you bad, jumped outta your chair on this one. And so I kind of have a feeling you might resonate with this.

No, um, I, I gotta throw out there, I was gonna, uh, before Catherine got into the, the policy thing, ask swearing, it's been a lot time, but yeah, the concept of an sbo om the software bill of material for the current release version of Adobe Acrobat as opposed to an SBO M four as we're look talking about here, some ephemeral web app that one time for five seconds exists in the cloud. You know, think about that. How do we, how do we deal with that?

Hmm. And I, and, but I think policy is, is the answer all hacking? All hacking is policy hacking.

I will figure out how you do things and I will figure out where the gaps are and I'll engineer that gap. And we live in a world right now where we generally have no idea what policy applies to any of us anywhere, with few exceptions. And in this topic, and because I'm used to the supply chain topic, imagine I needed to get the, the SBO or custody information about a piece of software on his phone.

Right now, I could get it in between five days and six months. Today I need to get it in half a second. That means I need to read the policies between me, the person who bought the phone and the first time the company I bought it from, and like their relationship, their contracts, their policies, you know, upstream all the way.

And we have to get that done in the next decade. So without unified and, and, and adaptable you know, transparent policy frameworks, none of this technology is gonna make a difference. So I think we, we will do that.

And there's interesting things going on down that path. It's kinda interesting in a way, just connecting dots between what you said, Catherine, and you were talking about Chris, there's your own unified policy management, right? Of what you're doing.

So, you know, you're, you, what you're applying where and how you're applying it. And then that's how that interconnects or interrelates with the people you connect with, work with, use their service product, whatever that is too. And I, and I appreciate what you said Chris, about, think about just serverless technology like a lamb to kind of service, right?

You know, it's there now, it's gone tomorrow may not be the same thing. It was a second ago when it, when it ran. Um, so it in some ways, Catherine, it's all sort of a dynamic unified policy management, right?

It can't be a static thing. Am I, am I on base here? Yes, of course.

You know, you do have to be responsive to the environment, um, you know, threat landscape. Um, this is one, one area where I strongly advocate for actually ML driven, um, detections and policy. Uh, this is a thing where, for example, if you have a really large data set, uh, you can train your ML models.

Um, how we do this at CloudFlare, just 'cause I think it's a little easier if I give an example and it's, uh, we will score each request on a scale of like one to 99. And if something is less than 30, that means like it is very likely to be an attack. And because we have, um, hundreds of terabytes of requests, or, sorry, hundreds of millions of requests every single day, um, we have so much data we could train this on and say a little blog in Malaysia gets attacked by a new attack we've never seen before.

Suddenly, because that tiny blog in Malaysia got attacked that gets feed and fed into our ML model. We don't have to rely on a security engineer to like go and find and analyze that attack and turn it into a regular expression like firewall rule. Um, the ML will basically just say, okay, like, since it matches something like this, um, we will just automatically block it.

And this is why I'd say ml um, sort of combined with that traditional, um, you know, security analyst looks at the traffic and writes a rule that matches it and then blocks traffic. Um, you gotta combine I think these types of approaches. So ML is a really, really great application, um, when it comes to being responsive to the threat landscape.

And we have some data around this as well. Um, we recently, not that recently, like half a year ago released our annual application security trends report. Um, and we found out that, uh, for example, like zero day vulnerabilities, um, we probably wouldn't have been able to find this out with just security engineers analyzing it.

But with our ml, we were able to detect, um, and exploit 22 minutes after the, uh, proof of concept was posted online. So, um, really, really great applications there. A lot of interesting stuff going on for sure.

Well, if that doesn't make the case for dynamic security, what does Right. Um, I, I'm curious, Kurt, h how do you, is is someone, you know, applying these things, applying security? Are you, are you looking at things like ml?

Are you doing it via yourself? It's something you look for in the vendors, the partners that you work with. How do you leveraging either that or the kind of technologies to help shorten that cycle between when things change and how you can account for it and secure it?

Right. Uh, I think the ML piece of it is, is hugely important because, I mean, humans, we're slow. The, the technologies we use, the computers, and I mean servers process all of this far faster than the human brain and I ever could.

And so we need to augment ourselves with this technology. So anytime we're evaluating new solutions and bringing them in, like I'm currently in the process of implementing a big one right now that focuses on platformization and ai, ml, it's all part of it because in humans with eyes on glass, like it's great to have those guys in the sock, but they'll get overwhelmed very easily with the speed at which things happen today. And so we need to leverage technology and machine learning enables us to do this faster than ever.

And it's only getting better, right? And so augment the human with that technology and you can very quickly pare down all of that information to what matters most and focus on real attacks like Catherine was just talking about. I wonder, you know, there's so much activity around ai, of course, a lot of it because of gen generative ai, um, Chris to, to security engineers have to become machine learning experts to be able to do this stuff.

What does it take to really leverage it? No, but knowing, knowing something isn't gonna have, um, uh, causing any problems. But, uh, I, I just couldn't agree more with, with both, uh, with Kurt and Catherine.

'cause you know, and, and your point, Kurt, it's all about time, time to transparency. How, how long, and again, I've seen this over and over in my career where we get to these points where what we're mostly doing is sharing the war stories. You know, I have no idea was 72 hours, none of us slept.

There was caffeine. And, and my my question always is, okay, if there was twice as much, what would you do? Because obviously that we're at the limit, we can't possibly work any harder to stay awake any longer.

And, and this, yeah, ai, ml, Oracles, whatever we call it in this, uh, my a big, there's been a big part of my, uh, my focus on supply chain before it would, you know, AI became, you know, uh, a general, um, uh, generative, what the hell do we call it? I'm sorry, I forgot. Yeah.

Generative ai. Yep. Generative ai.

Yes. Uh, too many terms to, To Throw around. Yeah.

Because again, we need to, you know, just for supply chain things, I need to read the contracts. I mean, I can literally call someone up, you know, it's not a security engineer, but it's some administrative person of the company, and I had to get them on the phone and get them to pull A-A-P-D-F and read the contract and find out if the clause allows me to get the information I need. That's not worth a human's time.

That's the kind of stuff that computers can do really well, and they're just beginning, but that's obviously the direction we're going. And if you can't see your policy environment five years from now, by various definitions, your competitors will be so much faster than you are that it won't matter anymore. Kurt, I'm, I'm curious, without giving us too many specifics about Teradata not asking you for that, but what's your sense of, what are the, what are the new priorities that are on your Yeah, on your horizon or things you're dealing with now and that you've kind of added in the last year or so?

What's changed about how you're thinking about security and that you've gotta address now? I think there's, there's always classic problems that we, we have to deal with and tackle. Like, we can't forget things like identity and network security and the rest of it.

But the, the prevalence in the emergence of generative AI and putting AI and machine learning in everyone's hands has meant that security teams have to be hyper aware more so than ever because these new technologies, people are latching onto them without considering the risks. Like, that's awesome. I can speed up everything I'm doing.

And suddenly you see a news story about, well, what was it like Samsung engineers leaked their code through regenerative AI solution or whatever. So you're, you can quickly lose intellectual property or put it at risk. And so we have to think about securing our environment for those solutions, or putting the guidance out for people to use AI and machine learning.

Um, and I mean, getting visibility of all of this, and another big one that's been getting pretty popular and we're seeing a lot from different vendors and acquisitions and whatever, is data security, posture management. Where is my data? Where is it moving?

How secure is it? Because at the end of the day, that's what the attackers want. They don't wanna sit in your network and use your resources to, to launch attacks as much as they used to.

They wanna grab your data, steal it, monetize it. So need, we're, we're focusing on data security big time in, in the more recent years, especially, um, forward looking because we have more data than ever. Interesting.

Catherine, from your perspective, you know, communicating with so many companies, what are some of the changing priorities from your, from your viewpoint? Yeah, I mean, certainly the gen ai, um, piece is something we're seeing a lot. Um, everybody wants to put an an LLM on their web application.

Um, and of course that means that you have to think of that as like a data security concern as well. Um, because you wanna make sure your LLM is not gonna like accidentally leak somebody else's social security number, because that's certainly happened before. Um, and so at, at CloudFlare we're thinking about this of like, basically how could you basically just put a WAF in front of an LLM, um, from that perspective, how could you prevent it from exposing sensitive data to the end user?

Um, but then, you know, you gotta think about these more complex issues as well. Like, how do you prevent somebody from poisoning the model? How do you prevent, um, you know, some of these other, like, how do you prevent it from hallucinating?

Uh, these are all, you know, sort of adjacent to security concerns. But, um, but nonetheless, we see some security teams focusing on this, um, increasingly. Um, additionally we also think about, you know, the, the LLM sort of security use case as a little bit of a just, um, increased API security use case since a lot of times, um, people are not building these LLMs themselves and hosting them themselves.

They're often, you know, bringing in LLMs from third parties, which, uh, necessitates, um, APIs, right, for integration. So how can you make sure that these APIs are staying secure and not leaking them back to the host and whatnot. Um, so that's definitely something we're seeing as well.

Um, I would say additionally, one thing I've been hearing a lot lately is, uh, software supply chain security. Um, I think Kurt mentioned the beginning, um, sort of securing code that lives on the client device as well. Um, this is something that we've been hearing a lot about, especially as it comes with the PCI four, um, compliance, which is gonna be mandated at the end of March, um, in a couple months.

Um, PCI four has a new compliance requirement around client side security and securing, um, the client side, like software supply chain. Um, so this is something we've been getting a lot of questions and inquiries lately. Um, you know, how much are organizations responsible for, um, the code that loads on their end user's devices, uh, when they visit their websites?

Um, this is something we are seeing a lot of people trying to actually actively get control over, um, and make sure that they're not, you know, serving, uh, code to the client devices that could do things like download a crypto mining software onto their phone, which, um, believe it or not, we have seen somebody's trying to make, you know, personal laptops part of a crypto mining network, which is pretty crazy. But, um, so yeah, I would say the client side component is, is something I've been hearing a lot lately as well. I, I just have to say, I, I love living in a world where we can use the term, uh, you know, hallucinating artificial intelligence in a conversation like this.

Seriously, just, we, we understand about that. It's Not a sci-fi movie. It's real.

Oh, It's, it's real. Yeah. Hey, so I've, I've kind of a left field question for you, Chris.

So if this, if I throw you too far off the track, I'm guessing you're thinking about this though, is, is there an SBO in our future for LLMs and s SLMs and all of these things? 'cause in a way, this is a whole nother part of the software supply chain, right? We're handing off to something that's doing inferencing, either on a chip on our handset or in the cloud, all of the above.

How does that fit into, do we need to be thinking or at least wondering how we're gonna solve this problem? And not only in not left field? That's, that's right in the middle of the, the, the tracks.

So in short, yes. You know, there's ano there's another assistant working group, um, Dmitri Rayman, uh, my colleague CTO at at sibe is, uh, a co-chair now one on AI bomb, right? An AI bomb has been talked about for a long time.

So what does that even mean? You know, so AI is code. So there's this, you know, same sort of standard SBO stuff about that, but there's also the training data and the models are produced, right?

And this sort of goes back to my last comment about ephemeral, ephemeral SBOs. You know, we start with the idea that IMA software provider, and every 16 years I release new code and I carve a new SBO m you know, on purist graphite. Um, but we live in a world where code gets compiled and used all over the place.

You know, how do we even look forward and say that I can commit to a policy that says I will, if asked, provide the contents of this code without, um, actually going out and printing or saving or producing quadrillions of SBOs forever, you know, in, in exabyte storage. Uh, so this AI is, you know, what we're currently calling AI is just another forcing function of the level of complexity we're at. So we need to be able to provide the answers to live up to the policies that we've agreed to, um, which is, you know, you know, in the SOM case we're talking about a software inventory that I will be able to tell you what code that was running or you know, what data set was used, and we have to get there.

And, and, and it's, it is reasonable progress down that path. It's a, it's a complicated one, but is very similar patterns to how we'll do other things of similar complexity. Um, Kurt is, is that on your radar yet at all, kind of thinking about security of, from a supply chain for LLMs and AI and ML algorithms and all that kind of stuff?

No, I mean, it's, it's certainly jumped up on the radar, especially since the whole SolarWinds thing happened. Um, as Chris was talking, it got the wheels streaming my mind of, well, if we're gonna be kind of, we're moving towards leveraging ai, AI in the sense and dynamically generating SBOs and things, is this another attack vector we potentially have to watch out for? Is how do you weaponize that and, and protect against it?

Because I mean, as we see attackers evolve their tactics and techniques faster than ever, they're coming up with new creative ways that defeat the traditional approach in microseconds. And so how, how do you stay ahead of that curve now? And so I obviously, I don't have the answer right now, but it's, it's really interesting as Chris talked to start thinking about this, this new sort of problem that we're facing.

And again, it all falls back to the rapid evolution of technology. Yeah. Speaking of that evolution, uh, just in the last week or so, uh, Satya Nadal, head of the Microsoft was talking about the death of SaaS, meaning that's kind the clickbait one-liner that what I think he was really talking about is e evolving nature of software architecture that I would describe it as.

Today's microservices or backend code are tomorrow's AI agents, right? We'll see more and more parts of apps built through, you know, with or through, or maybe completely with AI agents. And it reminds me of going into the, uh, cloud native era of, oh, how do we secure microservices now that we're gonna do that kind of thing?

That's kind of the, that's the next edge that we're, we have to work on and think about how, uh, there are different things we have to do for securing AI agents. How are they orchestrated? Is it Kubernetes or it's some other thing that's managing all those things.

And, uh, given that we're putting AI agent building capabilities in everybody's hands, in many cases, it, uh, could make for interesting. I use that in a nice way, uh, interesting environment to try to secure and manage. So in some ways, the future is bright, but it may be, uh, pretty intense at the same time, same time.

Well, and I think kind of building on that too is the, the technology behind ai, it's backed by machine learning. Like you're, you're making technology autonomous, right? So it's not as predictable anymore.

So how do you secure what, when you don't exactly know what turn it's gonna take next, Non-deterministic, right. Well, I, I gotta add a note, a note of hope, though, because it's easy, you know, to your point, uh, Kurt, the short answer is yes, because there's a new attack vector. Oh, yeah.

Um, but, you know, throughout my career I've been arguing this one, it's like, we'll probably keep the lights on. It's like, no, no, if we don't do this and that, then you, we will, you know, but the, that we're on this, we're doing this call right now. We've managed to figure out everything else over this point.

And not only that, but I think that we're, we've been mowing the lawn. I think, you know, what we need to do, generally speaking in cybersecurity has been known maybe forever, certainly 50 years, but we haven't gone around to doing the vast majority of it yet. 'cause we haven't had to.

But as we do, and I, I will take a risk and, and put a lot of my, my faith in policy, you know, and in real policy transparency, you know, in, again, in this decade, it gets harder to be an adversary because, you know, these are the happy World War II fans out there, you know, or no fans, you know, but the, the ubo wars, right? There was the happy days when you could just have a U-boat and sink shipping all day long. You know, that's kind of most of the world, most of the, the history of the internet to date.

It's not necessarily gonna just stay that way, that long forever, where there's always a new attack service, and there's always a, a, a new way when the last one is, is locked. I think we will, we'll keep it running. We will all be fine.

And I think over, you know, at least over a period of decades, being an attacker will become much, much more difficult. I mean, I might argue it already is becoming more difficult. It 'cause the, while, while the, the technologies we use as practitioners are getting more advanced, that helps make it more difficult for the adversaries of the world.

But that's not to say that they can't employ similar technologies, right? So now we're kind of, we're creating that chicken and egg problem all over again and playing a game of cat and mouth. It's kind of the next arms race, if you will, as technology evolves, everybody has access to it.

Well, let's do this. I appreciate all the conversation, and we brought up a number of topics, um, just as a kind of concluding thought. Uh, we, we've been talking about what are the things we need to be thinking about?

Maybe they're newer, maybe they're on the horizon, maybe already working on this today. Um, if you had to say, there's one thing you'd really want to emphasize this, if you were, you know, somebody who's listening to this and maybe making a few notes, the thing that sort stands out to you as something really important to be thinking about in the next, let's say, six to 12 months, if not today. Um, Kurt, do you want to give us your thoughts?

And then Kathleen, if you would, and Chris, you can wrap it up for us. Sorry, did I say Kathleen? I mean Catherine, excuse me.

Kathleen. I work with a Kathleen. Sorry, I've been doing that.

Oh, good. Okay. Yeah, I, go ahead, Kern.

I mean, it's, we wanna avoid that situation where everything is a priority, so nothing's a priority, right? I think we, throughout this conversation, we've highlighted the importance of esmo is we've highlighted the importance of application security and how it's, it's becoming more important than ever because our application code is, is literally going everywhere. And that's, that's kind of the gateway for a lot of the attacks we're seeing in the world today.

And so I think the, the emphasis is on application security, but it's also just say, let's not forget the rest of it, because all of the, the other parts of cybersecurity are hugely important. And we still need that visibility. We still need the coverage, and we need to be thinking about ease of use as well, and avoiding the sprawl.

So I know these aren't necessarily specific cybersecurity things, but they, they help you simplify your approach and, and focus on what matters. And that depends, that, that changes everywhere you go. Every enterprise or company has different priorities.

And so I think focusing on those things help enable us to, to focus on what matters for where we're at currently. You good, Katherine? Yeah.

So I mean, like Kurt said, you know, we wanna make sure that we're not making everything equal priority. So I think when it comes to application security, which is of course, my area, what I would say is most important in this space is visibility. Um, the attack surface is getting more complex, applications are getting more complex.

Um, you know, where they're hosted is getting more complex. So how do we actually have visibility into our entire, entire application attack service? How do we have visibility into the APIs developers are creating so we can actually secure them?

How do we have visibility into the software they're adding, um, to these apps? Uh, that I would say is probably the most important thing for application security and also one of the most challenging things. Excellent.

Chris, I, the, you know, Kurt and Catherine both want exactly where I'm going, so I'll just build on that. You know, do do things that save you time to transparency. You know, if you, you know, don't panic, nothing's on fire.

And, and when things are on fire, panic less, right? Just take your time and, uh, getting visibility, you know? Yeah.

Look at how long it takes you to figure out. And anytime you find a, a, a way, you know, in this, in this topic we're talking about here, to spend less time to figure things out, you have all that time back to do things. And it's easy to just, you know, particularly in transitional periods, to just do more and more and more of what you've been doing, you know?

But, uh, understanding the environment you're in so you can apply your resources appropriately is, is everything. And there are lots of ways to do that these days. You know, there's, there's a lot of Russian panic, and there are a lot of, you know, I will say it, AI and things like that out there who will actually make your life easier, give you some of your time back.

Mm-hmm. And you point feel better knowing what's going on, make and make better plans, that better strategy. Yeah.

To that point. Exactly. Chris, and, and Catherine mentioned it around, uh, ml, you, some of the things that I'm really excited about AI is actually just the understandability of what's happening.

You know, Kurt mentioned about as things ramped up or, or you did, uh, uh, in, in the, if the tax doubled, right, how would we handle that if we're already maxed out? So some of it is just handling the volume of things that are happening. But I think one of the things that I think is most exciting about generative AI is it's also so complex.

No one person can understand the full system, right? Or maybe even understand truly what's going on in a case of an attack or where you have vulnerabilities. And generative AI is starting to make some inroads and helping us understand systems and, and giving us some insights to some of the complexity.

We may not be able to fully get into our head all at once. So, for example, I've been doing some work around how do you modernize mainframe applications? Well, nobody was around that built those things.

Well, maybe people that built the network aren't even around, right? So help us understand what really is happening with all this data that we've collected. And the natural language interface through that is, is a great aid, and I think it's just a real practical thing that we can start to begin to use today.

So don't think of AI as just as the next, you know, it's gonna replace all of our software and it's all gonna be different. And what do we do? There's things today that is already helping us with.

So, you know, there's some real things too, not just what's on the horizon. Well, thanks to all of you. It's been great, Catherine.

Uh, we appreciate your perspective, and Kurt, you're bringing, um, your experience and perspective. And of course, Chris, always good to be chatting with you and your connections into the security world. And some of the folks are working, collaborating together, which by the way, is another superpower we have in security.

And that's the fact that we work together and collaborate on, on these things. We're not going at it alone. So thank everybody for the good work that we're doing to help advance.

We hope this has been a helpful conversation for you and thinking about the, the last great cloud transformation, what we're doing differently and thinking about, uh, as we move forward. So as we've got our heads down, getting stuff done, getting our priorities done, getting our plans in place, and executing for 2025, but also kind of thinking a little bit about what's next and what we might be considering and learning from others that are working in our space. So thanks to all of you.

Thanks everybody for joining us today. And thank you to the Cloud four team for, uh, for sponsoring, um, our show today. And we look forward to joining us either on another recording or Sure.

And check the calendar for one of our live events where folks can ask questions and engage with us in a similar kind of conversation. We have many of those coming up. We'll talk to you again soon.

Take care, everybody. This is Textron tv. Hey guys, thanks for the throw.

We're here with Dale Hoke, who's senior director of Information Security at Red Scale. And we're talking about a new continuous controls monitoring report put together by the CISO Society. And well, that really just goes to the whole heart of this, what's going on with GRC and cybersecurity.

Dale, welcome to shout. Thank you, Michael. I appreciate the invite.

Looking forward to having a conversation. So give us the high points of the report and why you guys think that this is significant, because, um, there's just a lot of things to worry about in the land of cybersecurity and compliance. So how do we make this a priority?

Well, I mean, the, the key factors around the report that kind of surprised me was 95% of CISOs don't believe that they have a good control on, on their, um, on their programs. And, uh, but outta that 95%, they all believe that that automation is the key, right? So when you start talking about GRC, I don't know about you, but in my head, I hear antiquated processes.

I, I hear, uh, spreadsheets, I hear right? Siloed data, I hear people who can't have conversations or data stored in, you know, disparate places all over the place. What I hear con continuous controls monitoring, right?

That is more of, you know, a dynamic operational control assessment, right? I have, I have a security tech stack, I take the outputs of that security tech stack, and I get good compliance, right? And I do that by optimizing my resources, right?

It's a force multiplier, really is how we look at it. How do you keep people from wasting time, right? How do you keep companies from wasting money?

I would say, you know, in the report it cited a good, um, 40%, uh, either didn't have funding for compliance or it wasn't a priority, right? Where I, I, I'm sure if you talk to some of these, uh, some of these, uh, auditors, right? They understand that, that in order for them to operate anymore, right?

Compliance is, is gonna be a priority. And how do we achieve that compliance without crippling security? That's the real question.

Everywhere I go, somebody's complaining about regulations, and yet it sounds like, you know, we're not doing anything to automate the process to make it less painful to execute. So are we involved in some sort of catch 22 here? Absolutely.

Right. So there's, I think there's three pieces of that. So first, nobody can agree on the standard, standard.

You have Osco that's been around, but it hasn't been used because nobody can agree that that standard is worth it. You have OCSF, right? Um, you, you know, the, the, that, that chosen framework should help you digitize your outputs from your security tech stack and bring 'em into generally any tool that you have out there to give you a single pane of glass into compliance.

But, um, you know, it's still subjective, right? So you have a subjective, um, deliverable, right? That, that somebody has to look at and agree with.

Then you have largely manual processes. So why would they spend the money to automate them if we can't even agree on a standard, uh, to report against them? So you see, you see NIST and, uh, FedRAMP is moving towards more of the compliances code NoCal, but I think we, we lose track of what compliances code is.

The digital outputs, the machine readability of these exports saves a lot of time. But also, you know, compliance is code the digital readability. And your CICD pipelines also operationalizes your security, right?

In my pipeline, I'm making sure that I'm building in security, then I make sure I build in, in compliance on top of security. And then at the end of that, I have GRC outcomes at the earliest onset because it is just too expensive to try to fix compliance problems or security problems at the tail end of a delivered product. Well, what is the relationship between the folks who are doing GRC and security these days?

'cause sometimes you hear about these two things are melding together, and then the next thing you'll hear is, you know, well, a compliance standard isn't good enough for security, it's just the bare minimum. So, um, how do we meld these functions? Or should we meld them in the first place?

Well, I, I believe that to answer your, to answer the last part of your question first, I believe you have to, right? I had think you, you have to meld them because if not, you're just gonna burn time and money, which no has enough of, right? Their people are doing a thousand things and nobody steals work, right?

They just keep stacking it on top of it. So they're just throwing more trash in on top of a dumpster fire and, and not putting it out. Um, I think what we need to do is improve, um, is look at, you know, your security tech stack should give you GRC outcomes, compliant outcomes, right?

You should be able to rely on the fact that my scanner is doing all of these things that needs to be done. And at the end of it, I have good reliable compliance because, Michael, I agree. Compliance does not equal security, but compliance was designed right to enforce security, and it should have been a roadmap on how to be more secure.

It's become subjected over time. It's, it is developed cascading problems, different verticals viewed risk management and compliance differently. And all of that has been, like I said before, just increasing the dumpster fire.

So I think we need to kind of take a breath, right? Recognize where I can get my GRC outcomes. I can take my scanner through the use of, uh, uh, you know, integrations, bring in my scan data, review that, scan data, digitize the compliance process, put it into a single plant paint of blast, so that not only CISOs understand the state of compliance in, in their environment, in their enterprise environment, they also understand the state of security.

So rather than having two siloed entities, which compliance people with security people, they did, I mean, they getting them to talk in any major organization is next to it possible. Um, that's just two of the biggest silos in this. And then you stack legal on top of that, or third party risk management or all of the other things that are coming about, uh, with CMMC and the, uh, you know, GDPR and all of the different, uh, cybersecurity frameworks that we have.

You just have a recipe for disaster. So you've gotta line that up. The only way that, that any compliance person is gonna line that up is by working with security practitioners, working with their operations team, developing what, developing what good operational security looks like, and then, um, getting compliance outcomes at the tail end of it.

And not using compliance as just a checklist, because that's not what it was intended for. That's just what it's turned into. So, as you kind of noodle this a little bit, I always felt like part of the exercise was to try to get those auditors in and out of the building as fast as possible.

And, but it seems like if the processes themselves are manual and the data's everywhere, doesn't that just increase the total cost of the audit? I, I believe so, right? So, uh, I believe that if you're, if you're not, you know, putting evidence collection on a timeline, this, here's what's happening.

You set up a continuous monitoring plan manually, you're assessing controls if they get assessed in accordance with the schedule, right? It, it, it's built into a three year process. And then six months before the audit, people are scrambling around, right?

That the auditors have gotta find the data, get the data, review the data. Um, the CM process should enable you, uh, if executed properly to have all that evidence audit ready all the time, right? Single pane of glass.

I know people laugh when you say that, but I'm a thorough believer in putting everything in one place where you can see it, right? And then putting a schedule over it so that it's pulled automatically, right? You don't have to reach out and wait for a week for somebody to open their email in order for you to get your evidence requirements.

It's already there. Oh, this is out of, uh, this is out of freshness, right? That you get a notification, it says, Hey, uh, you need to, you need to provide this evidence.

And it shouldn't take a compliance person, right? Sending an email or walking across the office to say, Hey, I need this scan, or, Hey, I need you to update this control implementation statement. That stuff should all be done, uh, through automatic notification.

And where you can, particularly on the technical controls, provide automated updates right there. It's really easy to update. Um, um, you know, your RAC data, it's easy to update any of the access controls.

It's easy to update, uh, any of the scan data, right? RA five should be, uh, entirely automated if you're working into a FedRAMP or, uh, uh, in this 800 environment. There's just, I could go on and on.

There's just so many places that you can minimize the pain points of an auditor by bringing that data in. It's there, it's, it's in a readable format that they're ready to go check, check, check down their checklist, and then they're out the door. So I believe it should, uh, and, and, and speaking with many three PAOs, they agree, right?

They don't want to be there, right? They, they don't, they don't, they, and no three vao that I've ever talked to wants to fail somebody in an audit. They want you to have a good plan.

And in order to have a good plan, if you have a good automation with good evidence collection and good policy reviews and workflows, that can all be optimized into a single pane of glass, which gives SSOs increased visibility and allows them to make decisions in near real time vice scrambling to prepare for an audit at the last second. Is this a psychological problem? And I ask it from this perspective, it feels like we historically, always thought of as an audit was an event that was gonna occur at a given time period.

And therefore we would get organized for the audit when it was coming. But it's really should be a process, right? I mean, it should be something that is just continuously running and it's not really an event.

I, I totally agree. I I believe it should be something that is planned for, you know, you're gonna get an audit. It shouldn't be a surprise, right?

Even a surprise audit, you should be audit ready, right? Through the use of the CCM practices. You should be audit ready all the time, regardless of, uh, that should be platform, if not agnostic.

That should be everybody's goal to be ready for an audit 24 7, 365, and to have everything in, in a workflow that you can produce evidence on, on demand. It's unfortunately, right? You get stuck in these silos where people are comfortable and the value of change is, is not recognized, uh, compared to the, the value of, of not changing, right?

So if you have these processes, they've been in place for a long time, people are comfortable for, you know, you know, every every company has a compliance guy that's been doing it for 30 years and has done it the same way for 30 years. So, quite frankly, automation scares 'em. And then add AI into the automation, right?

That's the latest evil word, right? So instead of recognizing that machine learning can actually shorten your time by about 80%, if not more, right? It's looked at as either one, you're gonna lose people, which nobody wants to do, or you're gonna lose control of processes, which these things are not true.

Every compliance analyst is overworked, every auditor is overworked. There's way too much work in order to meet the compliance standards. Uh, and without automation, without the use of ai, right?

I, I think we're just burning capital that no CIO has, And the audit itself, at least in my experience, is only true or current for approximately 10 minutes, because then somebody changed something and, and then we know we're back to the same cycle all over again where we're outta compliance and maybe there'll be a surprise audit or whatever. But to your point, didn't we just inject too much stress in this system and we're kind of our own worst enemies? I, I believe that, that that part of it is there, right?

I think part of your statement is true. I think that the other contributing factor there is that technology outpaces compliance, right? And compliance struggles to keep up, right?

So I, I believe that compliance is doing the best job it can. I believe that, that every AO is doing the best job that they can, uh, in light of the current situation. But let's face it, the drivers for technology, right?

The drivers for business practice against technology are going to supersede compliance, right? Because if you're telling a company that they have to, that they have to make money, right? Or they have to be compliant, right?

But very rarely the two line up, um, and where you, they can be accomplished EAs easily, the company's just get a fault on making money, right? So we as a, uh, we as security practitioners and compliance practitioners have got to find better ways to make compliance, keep pace with technology to improve business outcomes. And right now, right?

Without the use of any type of compliances, code automation, um, and, and industry without the, it's being used, but it's being used sparingly. So without that, that being put into place, you know, getting the good business outcomes and being compliant, it's very difficult. And the paradox here is that every nickel we spend on compliance is one nickel that we don't have to either bring back to the bottom line or invest in something else, right?

Correct. Right? And so, I mean, I think it was 46% of the companies, uh, did, or 36% of the companies, and, and I'm working off the top of my head here, didn't recognize compliance as a, as a business value.

And then 46 of them just didn't have the money. So, small companies, right, aren't gonna have the money to stay compliant all the time without automation, which we've seen in the report. Um, most of the companies that are embracing automation are small companies that need to be compliant in order to get into the marketplace, uh, and to make real money.

So it is, it is like a snake union its own tail, and it becomes really ugly. All right? So what's your best advice to folks then?

I mean, once that one thing that you just keep looking at and makes you shaking your head and go, folks, we need to be better than this Guys. It is not hard. It's not hard.

Find the places that hurt and automate them, right? And everybody has the same pain points if you sit and talk to 'em, right? You, you have a limited amount of experts or SMEs that operate in compliance and security.

You have the limited ability to stay on top of compliance frameworks and regulatory changes that happen. And you have a limited, uh, you have a finite amount of capital, you have a finite amount of money that you could throw at any problem. So find your top five paying points, right?

And, and do some research on how to automate 'em. And I think everybody's gonna, gonna come to the same place, automatically collect my vulnerability data, write it automatically to an issues tracking, whether it's a poam or a workflow or ServiceNow or Jira. Manage that automatically as it's fixed, my compliance is updated.

I always have good dynamic operational control assurance, and in a single pane of glass, I can see what's going on all the time, right? And this is not hard. A lot of people have done this right on the spreadsheet by writing scripts.

So now all of the, there's a a handful of platforms out there that'll take an automate this for you. Second, become familiar with what, uh, compliances code is, right? Get familiar with Osco.

There's a lot of resources on the NIST website. There's A-O-O-S-C-F is a, there's got a lot of good resources, right? Let's be, be familiar with what I'll call, you know, compliance benchmarking, where, um, these, instead of using compliance as a check mark, use the system on the front end to achieve appliance outcomes.

Um, and I do that by doing a business impact analysis, right? Recognize where compliance actually will improve your business. Because I think everybody, if they looked at it, honestly, would see that there's probably four or five spots that they could automate and save a bunch of money.

And then take your time and incrementally deploy it, and then go back and do metrics, do a metrics assessment. 'cause if you're not measuring it, it's not, it, it's not successful. So figure out how much time does my guy do it in Excel?

And then how much time do I do it after I automate work slow, right? It, it's, it's, it's, change is painful on everybody and it's scary, right? So don't, don't boil the ocean, right?

Chop the elephant up in the small pieces and eat it a small piece at a time. Pick your number one control that you think is the most painful, right? To me it's RA five, right?

Get your security scans in, automate the outputs, generate automated workflows to repair the issues, and then update your compliance and see where you stand at the end of that. And I think you're gonna find that you saved a bunch of money on your security team, your compliance team, and your operations team, all of which has to touch that workflow at one point or another. Alright, folks you heard in here, words to live by it.

If it's not fun, go find a machine to do it. Hey Dale, thanks for being on the show, Michael. I appreciate the invite.

If anybody has any questions, happy to answer 'em. All right, back to you guys in the studio. com is the leading resource for news analysis and education on challenges facing the cybersecurity industry.

com covers all aspects of cybersecurity, including data security, DevSecOps, cloud security, application security, network security, security threats, and more. com has the largest selection of security content featuring breaking news, blog posts, podcasts, and more. com to learn more.

com. Home of security bloggers network. Hello and welcome to the latest edition of the Techstrong AI video series Army host Mike Biard.

Today we're with David Brockwell, who's technical director for NCC group, and we're talking about AgTech AI and some of the security misconceptions that go with this next wave of AI technology that seems to be taking us all by storm. David, welcome to the show. Thank you so much.

Glad to be here. I'm not even sure that we all understand exactly what a AI agent is these days, but there are gonna be security implications. So walk us through what's going on here and how do we think about this in a way that maybe we can get in front of something for a change.

Yeah, for sure. Well, as we all know, AI has been the hot topic in technology and security for the past three or four years now. And ag agent seems to be the buzzword of 2025.

And there are a lot of misconceptions surrounding what AI is or what being an agent means. But overall, when we're talking about agentic ai, we're talking about the same systems that we've had for the past three years in terms of like chatbots, language models and so on. But the differences that we're equipping them with the ability to execute tasks autonomously on behalf of users.

So for instance, if I had a chat bot and I asked him, Hey, I need you to go out and purchase something for me, an agent would be able to come up with a plan of action and execute on that plan to fulfill my request instead of just giving me some sort of response. So What are the security implications of that? I mean, do I need to double check that the AI agent is behaving as intended or can somebody hack into that and just take it over?

Yeah, great question and one that I don't think enough organizations are really looking at right now. So at NCC group, we do security consulting and we've had the opportunity to be exposed to dozens and dozens of different implementations of AI agents. And one of the trends we've observed is that the more functionality and power you give to your agent to execute tasks, the more dangerous it becomes from a security vulnerability perspective.

'cause all of a sudden the impact of compromising an agent where before when it was a chat bot was limited to just giving the user some sort of negative response. Maybe you cuss them out or, or do something else that you wouldn't want it to do. Uh, but now you can actually cause them to execute a task on behalf of a user.

So I might be able to purchase products, I might be able to post, I might be able to manage accounts. And what organizations aren't thinking of a lot of the time is how do we properly manage access controls when these are still statistical systems? I mean, when it comes to traditional technology, unless some random cosmic flare changes the way that my program behaves, it's going to do the exact same thing every single time.

But with AI agents, if I as a threat actor am able to get some form of input into its context window, maybe it read something from my account, read something I put online, then I have the ability to manipulate its output and cause it to run tasks that it otherwise wouldn't have performed. If that's the case, then should I have a lot of smaller narrowly defined AI agents to make sure that the blast radius is limited? Or do I have one kind of Uber AI agent that is just extremely well protected?

Yeah, I think that when it comes to ai, we have to change the security model that we approach it. Having a monolith where we try to make everything surrounding the singular agent secure, uh, hasn't seemed to have panned out for organizations that have tried it. In fact, your suggestion is one of the most successful that I have seen, and it's a technique that I call dynamic capability shifting, where when an agent is executing on a task, when it needs to acquire data to run that task, its permission level is dropped so that the only thing it can execute is the narrowly defined set of capabilities it needs to fulfill the user's request.

So to, to give an example, let's say that I have an Amazon purchase bot and I ask it, Hey, I need to know the latest reviews for some product that I'm interested in purchasing. If I go and ask it to summarize those reviews, it trusts what I tell it to do because I'm the user. But it knows that it can't trust the reviews themselves.

So what the agent should do is the backend code, or well, first the agent should recognize that it needs to perform a summarization, which operates on untrusted data, tells the backend code that it's going to summarize the backend code, then drops the level of permission that the AI agent has to execute tasks as such that it can only summarize content. Then the agent gets that content, performs the summary and returns the response to the user. That way if I compromise the agent, it has nothing that it can do anyway with my account because its permissions have been dropped.

Aren't we gonna have the same problem then with the AI agents that we've had with humans where basically they're overprivileged and then people start escalating those privileges when they hack into them and then all hell breaks loose. So is this just a replay at a different level of scale? Yeah, that's a great observation and it's really what I like to call the intern in the middle scenario.

AI agents act much less like traditional software code and much more like an untrusted overconfident intern in our application architecture. So in the same way, if you wouldn't trust, and I apologize to any interns out there who are great, I'm sure you're wonderful, but, uh, if you as an organization wouldn't trust your intern with the keys to the kingdom, you wouldn't let them make purchase decisions, uh, wouldn't let them interface with highly important customers, you also shouldn't allow your agent to perform analogous tasks. So in the same way that if I as a manager told my intern, I need you to retrieve an invoice from a customer that I don't trust, uh, the intern probably shouldn't be facing that untrusted input.

Instead, they should go through a trusted middleman, like a comptroller who's behind the scenes handling all of those trusted operations. com/research that goes into architectures on how you can dynamically limit tho those kinds of, um, trusts that you give to models. And really organizations that aren't thinking about this in terms of how we can trust data dynamically are going to quickly miss the mark.

So ultimately, who should be in charge of securing these AI agencies, the security team? Or is it the business and analyst folks that create them that know what it is they're supposed to be doing? Ultimately, I think that we need to look at shifting security back with traditional code.

We often have a point and patch perspective or we point out a flaw or a chain of flaws, patch them, and then we're good to go. That that vulnerability is not there anymore. But with AI agents, we need to bake security into the design at the architectural phase of the application.

So I need to construct the way that my components interact with each other in such a way that the agent is untrusted by design or at the very least limited as much as possible. So in other words, I can't have like some sort of master monolithic agent that's running all of this and then later figure out, oh, well this is a problem and patch, you know, this prompt injection here or this piece of data that it gets there that is just untenable in the long run. Instead, we need to be looking at agents from a dynamic perspective such that whenever we're designing our application data flows, we see the agent receives a piece of untrusted data in this operation, therefore we need to change what it can do to match what that data needs.

How long do you think it will be before we see an AI agent get hacked? A lot of folks say in security, if you can imagine it's probably already been done, but um, where are we on this curve? Yeah, great question.

When it comes to proof of concepts, we're already seeing these in the wild from bug bounty hunters who have dug into some of these production applications by major organizations found flaws and reported them, and typically they're patched within a reasonable timeframe. But I anticipate that it won't be long before we start seeing prominent threat actors use these for truly malicious opportunities rather than just as an example to show their skills. So it wouldn't surprise me if within the next year we get our first major AI hack in the news.

And these hacks, as far as I understand it, the AI agents themselves are trained using LLMs, but they're built using traditional coding tools and all the dependencies that go with that. So if there's a flaw deep in some coding tool somewhere, won't the bad guys just use that to leverage their way up into essentially commandeering the AI agent? Yeah, and that's a great example of a supply chain attack that we've already seen be executed out in the wild.

To give you an example, on hugging face, they have countless repositories of different AI models that you can download. And so even if the model that I download itself is trustworthy, quote unquote in the sense that it operates at a decent level of performance, there's nothing stopping, well asterisk, nothing stopping that threat actor from embedding malware into the code used to run the model itself. So I could do something like give you a model whose weights contain what is known in in Python as uh, pickled code, and as soon as you download this model and run it on your system, it will give me code execution and I've compromised your environment.

And so there are a lot of different ways that we've seen threat actors compromise AI systems not from within the AI itself, but the software used to run the ai and that is easily as dangerous as any, as any flaw or vulnerability within the AI software itself. Well, you think there'll be regulations emerging around these AI agents as it pertains to security and governance, or do the people in Congress really understand what the, the level of conversation really is? Yeah, that's an interesting discussion to be had.

I think that both the Biden and the Trump administrations have proven themselves to be fairly, uh, bullish when it comes to ai. But in terms of security, I don't think that the industry itself really knows the direction that it needs to go much less those in our government who are making these legislative decisions. Ultimately, the message that I like to give people is that security when it comes to AI has not rewritten the fundamentals.

We're still dealing with the same types of security issues that we've seen for the past 30 or 40 years. Just the nature of how those are executed upon have changed in AI environments. So, as a result, I think we need to be thinking less about what specific AI regulations do we need to have that account for AI specific vulnerabilities, but rather, what legislation do we need to bring AI systems up to the same security standard that we've seen for other technologies in the past?

And how many AI agents are there likely to be in need of some level of security and governance? 'cause in my mind, I can imagine that I might have 10 for various tasks, and then if each person has 10, well, you know, you're getting into the thousands pretty quickly. Yeah, I mean, ultimately I think that every agent needs to be designed with security in mind.

And right now, when we do these types of assessments, I would say less than 5% have been structured with the requisite security considerations and guarantees that they need to have to be secure whenever they're deployed out into the wild. So what's your best advice to folks? Because there's a lot of security people that struggle with the same issue all the time, is the tech gets ahead, uh, people don't think through the security implications.

How do I get, insert myself, I guess, into that conversation in a way where I can have an impact now versus trying to clean it up a year from now? Yeah, great question. When it comes to ai, we need to change the way that we think about trust from just a component level perspective to one that incorporates the data itself that these systems are using.

So in other words, it like with the traditional application, I might say that as long as my dependencies are secure, and as long as my developers code is secure, then I don't need to worry about security vulnerabilities cropping up outside of maybe business logic. But when it comes to ai, in a sense, everything is business logic. So if I misplace an assumption about how I trust my ai, when it receives untrusted content, then I am to introduce security vulnerabilities into my system because that AI is no longer a trustworthy component within my environment.

So we need to think of trust being polluted downstream. So whenever an AI receives a piece of untrusted input, I need to consider the AI exactly as trusted as the input that it receives. So in the same way that I wouldn't trust an Amazon review to make changes to my user's account, I shouldn't trust an AI agent who reads Amazon reviews to make those same changes to my account as well.

Do you think that the bad guys are kind of sitting back and maybe having a little chuckle amongst themselves? 'cause essentially we are now exponentially about to increase the attack surface and they're like, and it's even better, they don't even have any idea what the security issues are. Yeah.

Well, somebody who does security consulting on the offensive side where my job is to look for security vulnerabilities, I think all of us get a little bit of a rush when some new grand technology comes out that takes the world by storm. Because the first thing on our minds is, okay, what's next? What can I go out and hunt?

But in the same way that we saw the same pattern with cloud, with blockchain, with internet of things, we're going to see the same cycle of the industry rushing to implement a new technology, doing it incorrectly, seeing a lot of security vulnerabilities, and then finally having those vulnerabilities patched by new security standards within the industry. So in the same way that I think threat actors probably are, to your point, really excited to see a new technology take the world by storm, I don't think that it's going to be something that inevitably cripples us. I, I do think that this is another part of the same cycle that we see every five years or so.

Well, let me ask you this then. Will it play out this way? Someone will create an AI agent and another person will create an AI agent for managing the security of that agent and for governing it, and the agents will kind of eventually, um, circle each other in a way that gives us better security because there are checks and balances.

Yeah. So almost, uh, turtles all the way down, or an agent or a Borrow, something like that. Yeah, You know, it's interesting.

I see a lot of organizations trying to approach it from that perspective. And I think that what we observe is that the more agents you put in in a chain, the more likely it is that one of 'em is going to pick up on the fact that, wait a second, something's going weird. Now, we should probably shut this down.

But in the end, AI itself, like I always say, is a statistical model. All we can do is strongly suggest that it follow the rules, and it just takes one threat actor to suggest even harder than our developers to get it to misbehave. So as a result, I don't recommend relying on AI to be the security control that protects ai.

I, that's what I would call a soft control. It's a defense in depth measure. Rather, we should putting, be putting in and implementing hard security controls from architectural perspective such that even if the AI wanted to misbehave, it would have no opportunity to because of the code surrounding it.

Alright, folks, you heard it here, no matter how you look at it, and no matter how awesome AI is, the more there is of anything, the harder it's gonna be to secure. So AI agents are no different. David, thanks for being on the show.

Thanks so much. Enjoyed it. All right.

And thank you for all watching the latest episode of the Techstrong AI video series. You can find this episode and others on our website. We invite you to check them all out.

Until then, we'll see you next time. Good day everyone, and welcome to the 5G Factor. I'm Ron Westfall, research director here at the FU and Group, and today it's a good one.

I'm joined by my esteemed colleague, Olivier Blanchard, a fellow research director and practice lead for our AI devices practice here at the Futurum Group. In fact, Olivier just recently returned from the Sam Unpacked event. And with that, I think we have some good insights on what's going on in the 5G and mobile ecosystem realms.

And with that also we'll be focusing on just that the things that have really jumped out of recent, you know, vintage, uh, that merit our attention. And so with that, Olivier, welcome back to the 5G Factor. How have you bear been bearing up since the last time you've been on the 5G Factor?

I can't even remember the last time I was on the 5G Factor. I've been, I've been dodging your invites for a while, not because I wanted to. I've just been on the road it seems like a lot more in the last six months than I have ever.

Um, so I've been, I've been doing well. It's ju you know, it's, uh, it's hard to keep track and, and to keep pace with, uh, all the changes in the industry and the hurricanes and, and all of the, everything that's happening all at once. So, uh, yeah, it's good to have a minute to to be back and, and catch up with You.

Yes, indeed. And you know, perfectly understandable and right, you are. It's been a very kinetic last, you know, well year plus really overall.

And, uh, and I think our first topic that we're kicking off has played a major role. Yes, AI again, however, this time, this is something that I think is really galvanizing, you know, attention. And hopefully by the time our recording is published, it'll still be somewhat relevant because matters are moving very quickly.

And that is, uh, just on January 20th, uh, China-based Deep Sea released R one. It's a reasoning model that basically is outperforming, you know, open AI's latest, uh, oh one model. And that is, uh, verified and, you know, various third party tests.

And so what is interesting here is apparently 150 researchers at a Chinese hedge fund, also known as Deep Seek, has out flanked the entire work of tens of thousands of engineers and basically almost the entire Western scientific community. And what is this just with a handful of modified Nvidia, uh, H 800 GPUs? Well, let's see.

Let's stay tuned on that. Uh, more likely at least, uh, from my perspective is that Deep Seq was trained on more than 50,000 H 100 GPUs. And while that is good, it doesn't automatically mean that the AO AI ecosystem apparently has reached a point that can train on materially less infrastructure.

And that means we still, despite this breakthrough, we'll probably need to continue our massive commitment to figuring out, uh, ways to optimize AI training as well as AI inferencing. And that is a naturally links to at attaining the main objective of artificial general intelligence or a GI. Now, a little more background, uh, there's a lot of threads here, but one I'm gonna focus on initially here to kick off the conversation is that according to Alexander Wang, the CEO of scale ai, he's the one that basically I think initiated the notion that, okay, what is going on here is that the Chinese have access Nvidia advanced GPUs on a wider scale than many people realize.

And the reality is that the Chinese labs, that they have more H one hundreds than many people think, and he added and shared that understanding, is that deep seek has about 50,000 H one hundreds, and they can really, uh, uh, can't talk about it because it's against the export restrictions that the US has put in place. And as a result, they have many more ships than, uh, folks, uh, really, uh, I fully understand. But that's changed, uh, obviously dramatically.

And Elon Musk, uh, also seconded the motion. Now, Nvidia, on the other hand, has insisted that deep seat is an excellent AI advancement and a purpose example of test time scaling. So this is an important technique now that will, I think, get a lot more attention because of what, uh, deep seek has basically achieved.

And, uh, in addition, uh, deep seeks work illustrates how new models can be created using this technique, leveraging widely available models and compute that is fully export control compliant, uh, emphasized nvidia. And so, uh, this means that interesting requirements, uh, that require significant NVIDIA GPUs and high performance networking will still be needed into the foreseeable future. And this aligns with what they see as that there's now three scaling laws, pre-training, post-training, and now test time scaling, uh, the newest one on the block.

So in effect, uh, Nvidia is adjusting that the deep seek used export compliant equivalence of an A MC gremlin and pacer and soup them up to the top line equivalents of Maseratis and Bugattis. I know it's a loose analogy, but it's like, wow, this is, uh, pretty remarkable that if they only use export compliant GPUs to achieve this. Well, I think there's a lot of countering out there.

And, uh, as a result, there's this mystery, like what are is the GPU cluster that Deepsea actually used? I don't think you're really gonna know for a little while, but this is definitely, I think, fueling all the speculation just how replicable is this? Just how cost efficient is this really?

But the outcome is like, okay, great. Now we have a AI training that could be done on a much more cost effective, much more energy efficient level then yes, and it's also open source space. That's good news for the rest of the AI ecosystem.

Uh, now officially, uh, Deepsea claims have used only about 2048 Nvidia H 800 ships, which, uh, to train the R one model. And that is was alongside the, uh, 10,000 older generation, a 100 GPUs that had already attained, uh, before the US uh, imposed export controls. Now all this, uh, I think is going toward NVIDIA's credibility on US export policy.

Now, naturally, NVIDIA has to claim that yes, that the, uh, chips that were used were export compliant. And that's understandable. However, I think as we dig deeper, and if we take, you know, the opinion of folks like the CEO of signal ai, uh, then I think that means that there's more going on here that can be officially disclosed.

Bottom line is that, you know, many of the actions that are backed by China's government, I think as we understand, you know, we've seen this with the Huawei for example, is that it constitutes what can be characterized as a giant psyop. And it's, in this case, China is attempting to create a little bit more uncertainty among the investment and tech community about AI's prospects in terms of it being, you know, primarily driven by, uh, US or Western technology knowhow. And as you can see, it did effectively ship the spotlight from, you know, the half trillion dollar project star, uh, uh, uh, um, uh, uh, which has, uh, its own set of, uh, major, uh, questions.

And so, uh, we have been, you know, really looking at, you know, how can we improve scaling laws and seeking efficiency with AI despite the deep seek breakthrough. And, you know, uh, again, we've been using open source, uh, rag fine tuning and forking capabilities, all these AI techniques to allow smaller models to be more performant. And we've already seen companies such as IBM, meta minstrel and others have stepped up and really have made, uh, I think, an impact on how this can be better achieved.

That is, you know, right size, large language models that are better performant. And, uh, with that, uh, Olivier, from your perspective, I know you've already, uh, provided some, I think, very valuable insight on this. And what is your take on what's going on here?

I know this is just one aspect here, but what else do you see going on here? Right. Well, I mean, that, that was a lot.

Um, I feel like the, the meme, you know, with a guy like this, and it's like he's got the, the board with all the red string just kind of connecting everything. Um, so I think that everything that you said is, is, is on points, uh, including the, uh, the, the, the, the theory or the hypothesis around Chinese ops. Although I, I would caution that, um, if, if we equate the AI race between US and China to any other arms race or space race in the past, the, the, the, the prevailing sort of tactic to, um, harm the other side, your opponent isn't to necessarily undermine credibility, uh, of their model.

It's to get them to spend a lot more resources chasing the same thing instead of trying to make them efficient. So by, by releasing deep seek, um, China, assuming that the Chinese government is involved in some way in this strategy and the timing and in the, uh, the, the tone and tenor of this release, uh, the, the, the objective for China should be to just let the US overspend on infrastructure and AI resources instead of taking this sort of more efficient scrappier approach that they took to achieve the same results with fewer resources. So I I, I take the, the notion that that China is trying to undermine US markets and US investments in AI with a grain of salt a little bit.

Um, but having said that, I, I think, I think most people are actually missing the point with deep six. I don't, I, I don't, I don't know that we're, we'll still be talking about deep seek in six months. I think that the, the larger trend, which is something that we've been talking about for the better part of the last year, is that there's been an evolution.

There's been a, a trend line in, uh, in AI training and AI inference that transcends deep sea, uh, and, and any other company like it that is very likely to pop up in the next, you know, six to 18 months. And it's this, the, we're we're the confluence of, uh, an increased and, and sort of systemic, um, improvements in the efficiency of the model on the training side, on the inference as well, but definitely on training, um, where, um, models that, that had to be trained on the cloud with massive resources behind them, massive inputs of power of chip and compute to, to train these models have become so much more efficient in the last year that what used to be only trainable in the cloud can now be trained on PCs and in some cases on mobile. So we're seeing these, these models sort of shrink and get, get faster and cheaper to train, uh, just by virtue of the fact that they're, they're, they're becoming more efficient.

On the other hand, we have this, this other element, which is the chips are getting more high performance as well, which is the whole value proposition between NVIDIA's Blackwell, right? You could use fewer GPUs to, to generate more outputs faster and more cheaply. So there's, there might be a slightly more upfront cost on, on the GPU side, but you need fewer of them to achieve the same results and few less power, less, less, uh, water, all of these other resources, less footprints, uh, data center-wide.

So we have these two efficiencies basically working together to make AI training a lot cheaper and a lot faster. This was happening already. It is gonna continue to happen.

Uh, it's, it's one of the things that is fueling the proliferation of AI processors and AI models from the cloud to the edge, and creating this more hybrid ecosystem of basically sort of like cloud to endpoint, uh, AI training and, and inference, uh, which, which I think is the reality that we'll be existing in, in, in, uh, as, as early as later this year. Uh, and we'll touch on that again when we get back to our coverage of the, uh, Samsung event last week, because there's actually an element that that plays into this. And so, um, the issue with, uh, with, um, stargates and the enormous investment numbers that we were talking about a week ago, and sort of like this injection of excitements in, um, in AI infrastructure specifically in the United States, seemed a little bit, um, I don't know, warts last week when the announcement was made, the, the impression that I had given what we just talked about and, and sort of validated by the deep seek announcement, uh, a few days ago.

And the, the sell off, uh, of, of US tech stocks in, uh, in the last, you know, 72 hours, um, is this, um, this is already happening. And, um, it, it's the, we don't need to spend trillions of dollars building massive data centers that are going to house millions of chips and, and, and servers and racks and, and require nuclear power plants and, and some kind of, you know, weird re-engineering of our water supply and our water management systems. We don't, we don't need to, to completely just stop everything that we're doing and build these massive projects, um, because the models and the chips are becoming more efficient and this federated distributed AI training and inference is already happening and the costs are getting lower.

So I think that what we're seeing last week is, um, let's go back two weeks ago, two weeks ago at CES Jensen, um, introduced sort of like his vision for the next phase of what NVIDIA's about, and we're gonna talk about Nvidia, because NVIDIA has sort of like the, the market power and the best position in terms of, of GPU infrastructure, um, and, and IP when it comes to training, um, AI at scale. Um, his model at CES, the introduction of Blackwell, all the things that he wanted to do, what he talked about in terms of, uh, training for automotive, training for industries, doing all these, these virtual sort of, um, you know, digital doubles and digital twins of the world and, and different environments, it's still valid. None, none of that has changed.

Um, but, um, it, it's the 2025 spend on this. And the reality of, of where technology is today in 2025 with 2025 chips, when 25 models, the, the entire 2025 current layer of, of AI solutions and IIP is absolutely not where we will be in 2028 and in 2030. And I think that Stargates misses the point in that it, it looks like, um, the type of spend that we would put together if we assumed that these efficiency advancements, these efficiency improvements were to stop this year, right?

Uh, so we're looking at bills for 2030 with budgets that reflects where we are today in 2025, not where we will be in 20 28, 20 30, 20 35. And so I think that they're, they're a little bit bloated. I think it's, it's warped expectations.

Um, I think the front end investments, uh, may be accurate. It, it may be the right number, uh, dollar wise, but I don't think it, it's, it's necessarily, I think it may be over-indexed on the data center side, is what I'm saying. I think that, um, taking a more holistic approach to the entire ecosystem of, uh, of research of chips all up and down the, uh, the, the AI value chain from basically wearables all the way up to the data center is, is a more realistic approach.

So I think that, that we're gonna end up spending a lot less on data centers and a lot more on production, on supply chain and on all of those middle layers to, to create a, a more sort of hybrid, uh, ubiquitous AI ecosystem. Um, and, and deep six's announcements. And obviously the reaction to the market, which I think was an overreaction, but that's a whole other story, is just one of several proof points along the way that the increasing the rapidly increasing efficiency of, of AI training and AI infrareds workloads for that matter, uh, and the, the, the sort of like diminishing curve of costs associated with that is, uh, is going to be a disruptive force in some of the calculations that we've had about how many chips we're gonna need to be able to, to, to power this new, um, ai uh, AI economy, right?

Um, and that's not necessarily a bad thing for Nvidia and for a MD and for everybody else's involved with this, but it, it, we should perhaps adjust these, uh, these massive numbers that we've been looking at and, and look at it more as a term in terms of, uh, diversification of chips. So if you're an Nvidia, for instance, don't just focus, or if you're looking at Nvidia for, uh, you know, in, in your investments, I'm not make giving any investment advice. Uh, so, so, um, don't take that for this, but if, if you're looking only at, at Nvidia for its data center GPUs, you're missing the point.

You need to look at Nvidia all up and down the value chain from, from PCs and devices and iot all the way up to, um, all the way up to the data center. And if you're looking at other companies like Qualcomm or Intel or a MD or, or, or media tech for instance, uh, you also need to look at those, those device layers and those intermediate layers where there's gonna be a lot of scale to deliver AI organically through this hybrid model, if that makes any sense. That was a lot.

Yeah. And this, uh, topic warrants that I think we're only gonna scratch the surface, but I think that, uh, our outstanding, uh, viewpoints there that you shared Olivier, uh, you know, first of all, you know, let's look at the big picture here. I agree this is not an existential threat to Nvidia by any stretch of the imagination.

And yes, their portfolio has diversified over the last few years. It's no longer, you know, about, you know, only, uh, GPU, so to speak. Or at least, you know, there's that perception.

It's, you know, the software and the services, it's definitely been diversified. And yeah, I think, uh, one important takeaway from CES among many was, you know, what, uh, Nvidia is doing with Cosmos that is, you know, taking real world video and applying it to AI training and capabilities so that you have, you know, these outcomes that can be very useful. And, uh, just that real world settings, you know, how to optimize, uh, say what's going on in the warehouse as an example, and, you know, advancing robotics and so forth.

And so, you know, a lot is going on here, but I, I think, uh, what is going on, uh, with a deep seek and, you know, uh, for that matter, uh, project Stargate, uh, I think, uh, that's important to note here is that it's like the throt thing of, you know, the AI hype cycle here. And to your point, I think, you know, okay, what about the half trillion dollars that's been allocated toward, uh, project Stargate? Uh, what are the implications of, you know, a lot more efficient ai, uh, capabilities throughout the AI ecosystem?

And this is a variation of Jevons paradox that is, if you make something a lot more efficient, that will actually increase more demand for it because it becomes more affordable and more broadly available. And to your point, yes, when it comes to, uh, hybrid ai, okay, a lot of the heavy lifting, uh, AI trading that's done in its GPU clusters and data centers, you know, throughout, you know, say hyperscaler, uh, networks, uh, a lot of this, uh, can now, uh, uh, be done in a more distributed basis in your edge data data centers and so forth. But also, uh, to your point about devices, yes, you know, the ai, uh, interesting, let alone some of the training can now be done at the outer edge.

And that includes, you know, devices that are powered by Qualcomm arm, uh, media tech and Apple, as well as, you know, across the, you know, the broader edge, you know, uh, including, you know, Broadcom, Marvell, Intel, a MD, you name it, that, okay, this doesn't necessarily mean you have to enlist Nvidia, GPU, uh, GPUs to do this, uh, more distributed edge, uh, training and inferencing, but it's really, I guess, uh, just that, uh, welcome news for the entire a AI ecosystem as well as for the other, uh, chip players out there. And, um, and on that note, let's, uh, now segue to the next topic, which is really AI at, on the device level. And as we saw at Samsung unpacked, which you were able to join, is that Samsung announced the, uh, galaxy S 25 Ultra.

It's, uh, galaxy S 25 plus and Galaxy S 25, all aimed at setting a new standard for true AI companionship with, uh, you know, basically our, you know, context to where, uh, mobile experience is. And I think that's a, a very poignant way to, you know, position this and market this at the onset. And what we saw is that Samsung introduced multimodal AI agents, and as a result, the Galaxy S 25 series is really a first step in Samsung's vision to change the way users interact, uh, with their phone, and also, you know, with the, the real world for that matter.

And so it's using the, uh, customized Snapdragon eight elite mobile platform to attain a lot of these, uh, uh, goals. And also the Galaxy chip set is, uh, designed to deliver, you know, greater on device processing power for Galaxy AI and, you know, improve camera range and so forth, and let alone galaxy's, uh, you know, uh, pro visual engine. So I'm gonna stop there because you were there.

Uh, you have, I know, in-depth, uh, uh, perspective on this. So what were some of your key takeaways from, you know, what Samsung announced with the new Galaxy S 25 Y? Right.

So it was, it was really interesting about, of, of all the, uh, of all the Samsung impact events that, that I could have gone to this, this was probably, uh, one of the more significance. And I think, um, somehow a lot of, of that significance got missed by the coverage of the event. So let, let me give you my, my take on this and why, why I think it's so important.

I've, I've been operating under the, the assumption, uh, as an analyst and as a, a practitioner of, you know, where AI is going and how, uh, I've been, uh, operating under the assumption that ultimately what we want is ubiquitous AI, device agnostic ai, where you walk into a room, it doesn't matter where you are, um, the, the devices that is nearest you and most capable of delivering the agentic AI experience that you're, you're, as a user, you're expecting, uh, and doing it the more efficiently is going to be the one to deliver it. So if I'm talking to an assistant and prompting it and saying, uh, I wonder what the weather is today, or, tell me, you know, what are my appointments this morning? Uh, it doesn't matter if it's my watch, my, my headphones, my smart glasses, my pc, my smart speaker, whatever it is, that's where it's gonna go.

And with, with a agent, AI specifically, one of the really interesting user experience advantages or value propositions is this hyper-personalization where your agents learn from you. They learn from your habits, they learn from your needs, from your patterns. They become sort of like your best friend.

They, they know what you want and in what, in what format, and what style and what speed before you know it. But definitely when you prompt it so they can anticipate your needs and, and adjust your environment, your workspace, your calendar, your, your shopping planning, all of it, uh, for you. Um, however, what that requires is a lot of device to cloud integration.

And, and I'm not gonna get into the, the specifics of, you know, how that needs to work from an architectural standpoint and an orchestration standpoint, but it's extremely complicated, and it requires some data to be on device to be cloned in the cloud and to be accessible both in the cloud and on device, uh, so that there's no, there's no lag between, you know, the prompt and the response. So you can have natural, uh, naturally times natural language conversations with, with your agent and your ai, uh, assist it, uh, without having to wait for a response or like, Hey, I'm thinking about it, gimme a second. Um, what Samsung did though is something very different from that, which is super, super interesting.

And at first I thought, okay, this is the wrong approach. And then the more I thought about it, the more I realized, wait a minute, this is actually smart. What Samsung did is they prioritized personalization and data security, and data integrity and, and safety, uh, by essentially moving a lot of those processes, the training and the inference to the device itself, and blocking it from the cloud.

And so essentially what they did is they have two parallel systems. They have, uh, uh, because they're Android devices, there's, there's Gemini, right, which is kind of like the Android device to cloud, uh, AI platform that remains untouched. So anything that you do with search, it's gonna go out for Gemini, and, and you have this, this normal integration, but all of the super personal stuff that it's learning about you, um, essentially sort of like the personality graph that's, uh, that an na agentic AI is, it needs to be able to build and train itself on in real time based on your needs is, is super personal.

And, uh, and Samsung made the decision to keep all of that on the device secure, no contact with the cloud, nobody else is gonna train on your data, nobody's collecting your data. Google is not collect, collecting that private data. It's on, on your device.

Not only that, but it's protected by their NOx security solution and it's post quantum encryption, which means that currently, at least in, in theory, I haven't validated this, uh, but, but based on on what Samsung is telling us, um, it's, it's not decryptable, uh, with quantum computers, which is, which is really nice. So pros and cons of that. Um, pros, obviously, data security, privates, uh, all of your data remains private.

And, and you have this, this, I think, added sense of, um, of privacy that you nor wouldn't normally have with a lot of AI products and, and solutions. Um, and it's tied directly to samsa. The, the, the con is that if you lose your device, you have to start all over again with all of that, uh, the agent training.

Um, but device to device, when you upgrade from S 25 to the S 26 or whatever the next generation is, you'll be able to do that. You just can't like, upload your, uh, that private data to the cloud. Um, but all of this is done because the, uh, the chip that they're using, the SOC, the system on ship is, uh, is capable of doing that.

And in this case, it happens to be Snap or Qualcomm's Snapdragon eight, uh, elites, which is the, the sort of like flagship platform that, that Qualcomm introduced, uh, at their summits back in October, um, of last year. So it's, it's the latest and greatest, but it is a custom chip made specifically for Samsung. So it's, it's actually the Snapdragon a eight elites for Galaxy, uh, which has a few additional bells and whistles for, you know, some camera, uh, improvements and also for this, uh, this enhanced, um, uh, agent AI on device capability.

But it's, it sort of illustrates, I think the, um, the, the, the new paradigm of this distributed, uh, agentic AI where sure, you can do, you can train a lot of models and, and do a lot of things in the cloud, and there are things that work best in the cloud, and that should be operating that way just as a cloud service. But there is also, oh, All right, I'm gonna continue it, it paused for a second. Um, that'll be a nice place to cut.

But there's also a, um, uh, a, a, a really huge leap forward in capabilities of these, these ARM-based chips that are on mobile phones, that are on PCs that increasingly are showing up in, in smart watches as well, uh, in smart glasses, uh, in, in essentially every digital product that, that we touch. Um, there's more and more capable of, of doing this. And, you know, the, the size of the, the models that you can train and, uh, and, and do inference workloads with on a PC versus mobile versus smart glasses depends a little bit, first of all, on, on the, the system on chip itself, but also on the size and the form factor, right?

You can put a lot more processing power in a PC than you can in a phone, and you can put a lot more processing power in a phone than you can with smart glasses or on a smart watch. Um, but you also have to think about how all these devices work together and how they can pool their processing resources so that instead of processing something on the watch, you're processing some of it on the watch, some of it on the phone, some of it on the pc. Uh, and a platform like Snapdragon, which is in all of these devices, might be able to just kind of work altogether more efficiently than cross platform, uh, combinations to give a user, um, uh, an enhanced AI experience that is primarily on device or that can sort of separate, um, the needs of, you know, pushing some of the, uh, the workloads to cloud services and then keeping some of them private for a variety of reasons for speed and efficiency, for power efficiency, for cost efficiency, but also for privacy.

And so it has implications for, uh, consumers, right? You and me just wanted to keep the private, uh, and cheap and not having to pay for all of this cloud inference stuff. Um, but also on the commercial side for businesses, because now the more data they can, they can have in-house, the more secure their data might be, um, and the more processing they can do in-house, also, the lower the costs, uh, of, uh, I mean, they don't necessarily have to pay for as many, uh, instances of, of training or workloads that they're pushing out to a cloud through cloud service.

They can keep a lot of that stuff in-house. Yep. Yeah, I, I think it's good news for say, the health monitoring use case, this, and it aligns.

And that's, I think, one constant theme. Uh, when we saw, you know, with the launch of Project Surrogate, Larry Ellison step up and say, Hey, this AI investment has warranted because of advances that could be made in medical research or, you know, tracking medical records and so forth with that privacy respected, built in and so forth. And likewise, you know, just being able to have a device that somebody can have confidence, it's gonna maintain my privacy, but it's also can just do that, you know, detect issues, uh, before, you know, they get outta control or, you know, uh, just across the board any health, you know, benefits, uh, out there.

So that I think is, is certainly the good news. That was one of the, the, the main use cases that, um, I'm not even sure that it was, you know, I think it might have been under indexed a little bit at the announcement itself, but in the pre briefings, uh, we spent a lot of time on, um, on that particular use case, how, uh, Samsung specifically. But I think as a concept, having all of that data and that agentic ai, um, processing and analyzing and recommendation engine on the device as opposed to in the cloud, first of all is more immediate.

But also, um, I think if, if, and I think especially with medical issues, um, I think people want their data to be protected. And we've all been burned so many times with hacks and, and, you know, our, our private data, especially medical data, getting out into the wrong hands or just getting out in the public, this, this takes care of that. And so you have a, a personal assistant that is able to, through the use of different devices and sensors, whether it's a, a Galaxy watch or a ring or, or other sensors, um, e essentially guide you and give you feedback on how well you're sleeping, for instance, your eating cycles, what your calori intake is.

Uh, we were talking about things also about a more granular approach to your diet, which you can sort of enter manually, but also, um, the AI knows what you're eating, and so it can sort of extrapolate the types of nutrients that you're getting and not getting. So it's, it's creating and painting a, a, a much more, uh, again, granular and, and detailed and complex picture of your overall health, and it is intelligent enough and understand your cycles, understand your patterns, understand cause and effects of good and bad behaviors, make recommendations, um, monitor your health. And it's, it's, it's amazing because it's all on the device and it's all secured.

And, um, I love this, it, it, it moves some of the, the, I think the, the real true immediate benefits of AgTech AI out of the cloud in areas where it doesn't need to be in the cloud. Um, so some things need to be in the cloud, some things shouldn't be, and it, it's, we, we now have the capability of, of parsing that, uh, according to our needs and according to our preferences. And that's gonna, that's gonna change that, that changes the equation a little bit for, uh, for how we think about AI investments.

And again, it, it speaks to that diversification of, you know, focusing on, on cloud AI training and inference, but also focusing on this, these edge use cases that are becoming much more prevalent and with a, a huge potential for, uh, for adoption. And the, the footprint. If you look at the mobile industry, if you look at the PC industry, um, even if, if those numbers, uh, essentially the install base doesn't really grow that much, there's a refresh cycle here, uh, that's gonna bring a lot of these AI chips to this, this broad install base.

So the market doesn't actually have to grow to show results. It's the refresh cycle that's gonna show those results and, and push a lot of these, uh, uh, these AI chip numbers out. Um, and that's what I'm hopeful about.

Yeah, and I think, uh, that's a great segue for who else can, uh, benefit from, you know, the, uh, the potential large s uh, the AI ecosystem, uh, overall, uh, growth. And this ties directly to, well, you know, 5G service providers, uh, certainly, you know, uh, the folks who provide mobile services to, uh, consumers and, uh, businesses, uh, and specifically here in the us. And what, uh, I think is interesting is that we're seeing, you know, major US operators such as at and t and Verizon looking into, okay, how can we play a more integral role and monetize, you know, AI in terms of, you know, uh, being closely linked to the services they provide.

It's not, uh, exclusively mobile, but it certainly includes, uh, for example, their fiber services, business services. But I think what's interesting here is that they have the real estate that we touched on, Olivier, how do we push more AI capabilities, training and inferencing closer, uh, to, you know, where the customer is, you know, bringing the AI to where the data is is certainly the mantra that, uh, we've have heard a great deal about. And so what's interesting is that now we see, uh, that Verizon business has, uh, revealed basically a, a bundle of products that are designed, uh, for enterprises and also cloud providers as well as hyperscalers to, again, deploy those AI workloads at scale by, you know, basically offering a single platform, uh, that is designed to do just that.

And what it is, it's called Verizon AI Connect, and it's offering a blend of really the operator's fiber infrastructure along with its power space and cooling capabilities, uh, and with those resources using it, uh, to deliver it, but also it's backed by its virtualized 5G programmable network to really make the AI workloads, um, more, I, I guess you'd say, customized to the, uh, customer needs, uh, out there. And so what's interesting is that already, uh, Google Cloud and Meta have already onboarded meta platforms specifically, and I think that's important because, uh, these are, you know, going to be logically the early adopters of AI infrastructure, uh, platforms such as this that are offered by a, a major, uh, service provider. And, uh, not to mention Verizon and Google are also looking at ways to, you know, advance AI services for, you know, things like network maintenance as well as anomaly detection.

So this is in play, uh, this is something that I think will kind of help the, uh, service providers. Okay. Finally, uh, we have a way to use our real estate, uh, to take advantage of, you know, ai.

And this is not going to be necessarily the repeat of things like, you know, mobile edge computing and so forth. There's still that risk, you know, the operators still not, not be able to figure out how can we be integral to this. But I think, uh, this is demonstrating that, uh, they're off to a decent start and not to be outdone, a TT recently secured, uh, $850 million from the sale and leaseback of its underused, uh, co facilities, uh, from, uh, a property company, uh, capital rain as part of its copper retirement plan.

And so, uh, this deal closed in January, it involves the asset transfer of 74 properties across the US spanning more than 13 million square feet. So the bottom line here is that the operators are becoming, uh, smarter about how they can take advantage of the real estate assets that they do have today, you know, that is retiring co assets as well as other edge infrastructure and, you know, working with, you know, the, uh, the major, uh, AI players out there, that's IE the hyperscalers to, you know, basically come up with mutually beneficial ways to monetize AI in the near future and certainly longer term. And, you know, Libby, what are your thoughts on this?

Do you see the service providers really being able to step up and actually, you know, playing a meaningful role in this? Or is this something that, okay, it's another missed opportunity and the service providers will be reduced to kind of a commodity like provider of the infrastructure for the AI services that are running across, you know, the, the clouds out there? Uh, maybe a little bit of both.

Um, so yeah, no, I think, I think it's smart for 'em to do it. So, uh, you know, the, the situation that we're in is, uh, we're expecting, obviously if we're spending, we're looking to spend half a trillion dollars on building data center, um, infrastructure. It means that we're, we're looking for a lot more computes to come from somewhere.

Uh, a a lot of these, you know, you know, builds are, are years down the road. We, we can't really start yet. So, um, the expectation is that we're gonna need a lot more compute power very quickly.

Where can we find it? And if you have data centers already out there that are underutilized, right? Um, or that can be sort of, you know, a lot of that processing think power can be retested for, uh, you know, higher price or, or more premium services, um, it makes sense for a business to go for that, for that ROI, right, for that opportunity.

So one, if, if a lot of those data center resources are underutilized or not used at all, suddenly they can be, you know, assigned to this, if we can charge a little extra for this, because AI is more valuable than, you know, being in a 5G network, um, maybe there's, there's value in that as well. I think though, that, that you're right, ultimately there's, whether it's successful or not, that that is gonna depend on them and how they package it. And, and it, there are a lot of variables there.

I think at some point, those data centers and those resources age out, they're no longer, um, the most efficient, the most cost efficient. Uh, and, and maybe they just get retired or the Verizons and at and ts of the world upgrade their systems specifically for those AI workloads. And now we start seeing a, a, a transition of spend where they also become, you know, they're, they're buying the black well powered, uh, racks and, and, and they're becoming more AI focused, and they become part of this sort of like, uh, ecosystem of, uh, of, of AI workload, uh, training and inference services, and it's all interwoven.

Maybe that's, that's possible, but for right now, at least for the next few years, while we wait for all of these, these massive data center builds, um, they're there with capacity. And so absolutely they should go after that market and, and see what they can make of it. Um, worst case scenario, it doesn't work at all.

Middle best case scenario, uh, they make some money, it becomes commoditized, and eventually it just kind of dies out because, uh, they're outperformed by other outfits. And best case scenario, they build a whole new business model that's, uh, that's gonna be really lucrative for them. So it's worse the shot.

Yes. And I think, uh, this is again, you know, AI and it's open-ended possibilities. Uh, there's just a myriad of variables here.

So on the one hand, it's good news for the sa uh, service providers. Certainly a test time scaling, uh, is, is introduced to the possibility of, okay, AI as we do it could definitely become more commoditized, as you pointed out. And thus, you know, that's good news for the, uh, service providers that this become a lot less expensive to, you know, do the, uh, AI infrastructure hosting and so forth.

Uh, now they, however, is, uh, will this result in monetize outcomes for them? And so, yeah, that is again, that kind of, uh, one of the many, uh, uh, uh, I would say the three doors that you presented, uh, possibilities. And it's hard to bet right now on the operators because the entire a i ecosystem is basically going through a lot of flux as we speak.

Yeah. And so, uh, we'll, uh, we'll come back to this. We'll talk more about it.

Yeah. Yeah. We should, we should come back to this six months from now and see, uh, Uh, definitely.

Or let alone yeah, say after Mobile World Congress, uh, you know, a lot of changes from that. Um, so this has been great. Uh, thank you so much Olivier, for coming on board and again, appreciate, uh, the, uh, opportunity for you to share your thoughts.

Yeah, thanks for, thanks for having me on. And I am not going to Mobile World Congress as of now, this year. I'm skipping it.

I'll probably go next year. Um, but, uh, but I'll be, I'll be looking forward to, uh, to the announcements surrounding MWC. 'cause I'm, I'm sure this, this very topic is gonna be, uh, is gonna be one of the major, uh, themes of, uh, of the trade show Yeah, Yeah.

That we can bet on. And that's something I think we kind all agree on. A and might as well call it AI World Congress for the time.

B, but, uh, and, uh, well, great. Yes, uh, I know, uh, we'll certainly be sharing, uh, thoughts on Mobile World Congress and, uh, and beyond that, uh, thank you everyone for joining the 5G Factor. Again, you can bookmark us on the future and group, uh, website as well as we can be, uh, viewed on tech strong, uh, tv.

And, uh, we certainly, uh, again, appreciates, uh, taking the time to listen to our thoughts. And with that, everybody have a great AI and to test time scaling as let alone 5G day. Again, thank you all.