MCP and the Future of Cybersecurity | Security Boulevard Ep. 3

We're actually gonna be talking about MCP. And if you went and watched Tron over the weekend, you may have been one of only eight people that did. However, we're not talking about the master control program.

This time we're talking about something different. Model, context, protocol. Welcome to the Security Boulevard podcast, a cybersecurity podcast from the Futurum Group.

Each episode explores a variety of topics within cybersecurity and the technologies that drive it. com, the Security Boulevard, YouTube Channel, tech Strong tv, and all of your favorite podcast platforms. Before we jump into today's episode, let's meet the panel starting with my friend Alan.

Alan, welcome back, Tom. Thank you. You know, this is my first, uh, appearance on the New Security Boulevard podcast.

We, of course, we've been podcasting around the Security Boulevard name since we launched it. Uh, I'm Alan Shimel. com Cloud native.

com, as well as Techstrong tv, and all of its various frontiers on the, on the rich media landscape. Um, beyond that, though, I've been in security for about 30 years and started several companies, co-founded several companies in the, we used call it InfoSec space before it was cyber. And, uh, it, it's a passion of mine, so I'm happy to be here.

Thanks, Tom. Well, Well, we're happy to have you here, Alan, as well as our special guest for this episode. Mr.

Scott Roon. Scott, tell everybody who you are. Hey, Tom and Alan, great to be with you both.

I'm Scott Roon. I am a happy and grateful, uh, serial Tech Field Day delegate. Looking forward to a couple events coming up in the next few weeks.

I am a co-founder of the Network Automation Forum, um, and I'm the co-founder and CEO of my consulting company solution, where we help people rationalize AI adoption and kind of cut through the hype, Right? And of course, I'm Tom Hollingsworth, event lead for Security Field Day and other exciting events within the Tech Field Day business unit at Futurum Group, and one of Alan's coworkers. I did wanna call out that it is, uh, gonna be National Dessert Day.

So, uh, hopefully you guys are having a sweet treat while you're listening to us. Um, and hopefully it has real sugar in it 'cause it's also National Real Sugar Day. Who knew that?

Um, well, the sugar, sugar people did. Uh, but I wanna talk about a topic today that's actually kind of sweet and that has to do with ai, but hold on, hold on. Before you race to the comments to tell me that we're talking about AI again, we're actually gonna be talking about MCP.

And if you went and watched Tron over the weekend, you may have been one of only eight people that did. However, we're not talking about the master control program. This time we're talking about something different model context protocol.

It's the hot new topic this year that has really driven adoption of a lot of different age agentic AI components, but of course the Security Boulevard. So, you know what we're gonna talk about, we're gonna talk about the security aspects of it, and Scott, you kind of jumped in with both feet on this, uh, topic. So I maybe you could kind of introduce to the folks out there.

What is it about MCP that makes everybody want to use it, and why is it the security parts of it are starting to maybe keep you up at night? Well, to, to give kudos, right? MC P's been with us for 11 months, you know, at the time of this recording.

It's really just kind of hit, hit the street, um, very quickly. I, I'm a networker who has done a lot of network security stuff in my life. And to see yet another product come out so quickly, you know, that wasn't governed by let's just say very long processes and, um, other, um, standards bodies.

It's, it's pretty amazing to see it. Um, but it is reflective of this typical, um, product life cycle where step one is get it to work. Step two is find a way to manage it, and then, you know, sometime later, step 179, secure it, right?

So there's like, let's make the packets go, um, and then we'll figure out how to secure all this later. Um, I have been pleasantly surprised with like, the integration of OAuth into MCP already. That's a great step and, you know, makes me hopeful in the first, you know, year of its lifetime here.

But, you know, miles to go before we sleep on this. So it's very interesting that you talk about this idea that we need to make it work first and we need to fix it, secure it later. I've heard this before, and Alan's the one who actually brought this up when we were kind of talking about it pre-show, uh, Alan, this feels a whole lot like APIs to me.

Is that what it kind of feels like to you? Well, There is a resemblance, but, you know, invent it, make it run, worry about security later. That's pretty much standard operating procedure in IT for the 35 years I've been here.

Um, you know, it, it's always the caboose on that train and it's kind of the afterthought, Tom, I gotta just one personal pet peeve of mine, I gotta throw out there. I hate all of these made up holidays and days. I don't know if hallmark's behind it or what, but it, it's, it, I think it's been the bane of it.

It, it's what led us down a slippery slope. So I don't care about National Sweet Day, sugar Day or any of those days, but back to security, this is a typical IT innovation. Wow, we got the greatest thing here since sliced bread.

Let's go out, run as fast as we can and get as much as we can. And in this age of ai, time is condensed so things are happening faster, quicker, speedier, more velocity. And then, you know, someone says, but well, but wait, but wait, what, what about the security?

Scott? I would, I would posit that zero auth wasn't put in for security was put in to make it easy to log into these things. Sure, yeah.

You there? That that is certainly an element to it. I'm gonna take a small win here where I can, but I I totally hear you.

Yeah, I mean, Tom, back, back to your point though, yes, this, to me, this was very analogous to the whole API security thing because you know, what we're really talking about with these MCP servers and, and you're right, Scott, I, you know, I thought it came out around last January, so I have 10 months, but maybe it was 11 in, in 10 or 11 months. It has become the defacto standard for how agents talk with agents and other agents with, you know, within your enterprise. Now, there is a competing standards sort of competing called A to a, which is the, under the auspices of the Linux Foundation.

And maybe that'll have better security, I don't know. But quickly, this is the defacto standard for agent to agent communication. And, you know, God bless the age of AI in, in, in 10 or 11 months, this thing is like, it, it's like it's been around forever.

And, and, and we had the same thing three or four years ago. You know, someone came up with the, the API economy and, and, and the studies were done that a majority of the traffic going over the net over the internet was API to API traffic web, web to web kind of thing. And all of a sudden someone said, you know, the most important attack surface left in security APIs.

We need to, we need to know what APIs we have, what their configurations are, what their security posture is. And until we have that, we're, we have a blind spot to a majority of the traffic on the internet. Scott, you're you're in the space, you know, you've heard, you've heard the stories.

Yeah, I think so. I couldn't agree with you more. Right?

Um, and some of the work we've done in network automation forum, uh, we put up a blog post a few months ago, that's just a collection of community comments, not necessarily an API security, but API quality and usability. And we definitely caught a thread of, you know, there's a lot of inconsistent, you know, publishing of API specs people, you know, getting half functionality, three quarters functionality, and that being a suppressor to actually driving automation for network and security equipment. So there's room to improve there for sure.

Maybe that's more of an issue in the network element world where they're trying to be more API forward and move away from, you know, the CLI addiction, that so many of us, especially Tom still have, sorry, Tom. Um, you know, but there's, there's gotta be room for improvement there, you know, from a security and functionality and consistency perspective. So let me offer this, because I think you're both right, that sometimes it feels like the security that we get with these rapidly developing protocols is more happenstance than it is by design.

You know, uh, you know, Alan, you, you brought up the fact that, you know, using authentication mechanisms that just happen to be secure to ease, user login may have had a serendipitous benefit. What about, you know, using h TT PS calls for APIs? Like, I'm sure that that wasn't top of mind for them originally.

Like, oh, yeah, you know, we'll just do this or we'll do that. Oh, hey, it just so happens that we've secured this communications channel by default years ago because everybody on the internet did. And when you look at the way that a lot of, especially things like electron apps that are basically just web browser wrappers anyway, um, we've, we've arrived at a state of security that is maybe a little bit harder to configure on the back end to begin with, but ultimately is more secure in the long run because the fundamentals of what we've done have changed so significantly.

Do you Feel like, go ahead. No, no. Greg, finish your thought, Tom.

I I, I was just gonna say, do, do you feel like that the people kind of behind the scenes that have been working diligently to do this have done more to bring the state of security forward than those people that are living on the bleeding edge? Yeah. So I, I will tell you the, the, the, the, the big issue in API security is most organizations didn't even know what APIs they had open, what APIs and what APIs were talking to, which APIs.

And you can't defend something you don't even know you have, right? Yes. The communication, you know, via H-C-T-P-S protocol and, and, and, and in transit is great, but if you don't know what you have and where it's going, it's good to know whether at least while it's going, it's encrypted.

But you still gotta, I would, you still have a fundamental issue. Um, that, that being said, though, funny thing happened with API security. There was a whole cottage more than a cottage industry.

There was a whole bevy of API security companies that came forward. The big one was called no Name. And then there was another one that got, I believe, bought by Red Hat.

And then there was, there was a half a dozen of these, uh, trace traceable io ai, excuse me, traceable AI was another part of Harness. Now, as often happens in technology, and Tom and Scott, you've probably both seen this, today's product became tomorrow's feature. Sure.

And so, API security, including API discovery, API communication encryption, API configuration has now been subsumed into some of these, you know, platforms that, that we, that we have around security. You know, it's, it's funny. I'm sorry, finish your thought.

No, No, that I, that was my thought. So, um, I've had the real honor of doing some work, um, associated with DARPA through something called the Embedded Entrepreneur Initiative, where, you know, DARPA works with lots of startups in the security space and other technologies that come up with really interesting piles of tech. Um, I won't call 'em products, right?

But they come up with fundamental artifacts and pieces of functionality, and they need help figuring out, should this be a standalone product? Should this be a feature on somebody else's platform? Or you're too late with this.

It's already, it already exists as a product or a feature on another platform. Um, but it's been amazing to see those elemental, you know, uh, I, I did some really interesting work on, um, IAM um, issues in graph theory with a particular company and said, try to figure out, um, does this belong somewhere? Turns out Wiz already had that covered.

Um, but, uh, yeah, the integration piece is really important here. You know, many companies come along with pieces of fundamental technology and how well they're integrated into bigger platforms, bigger products matters. Um, and, you know, we see it, we see it in the, um, network automation space.

Like there's really cooperative, um, open working together between lots of different technology providers. Um, how well you do those integrations really matters on how secure it is in the long run and what ease of use looks like. Yeah.

Let me, and I don't mean to scare anyone out here though, but let me frame the, the real problem we're dealing with though, right? APIs and APIs, talking to APIs were great. I think as we sit here now, on the dawn of the AI age, or in the early days of this AI era, it almo the automation that API to API gives us almost seems quaint because it was automation without autonomy.

As we enter agentic ai and what does agentic AI really mean? It means we're gonna have autonomous agents who aren't just doing a task. They're creating, they're, they're completing missions, multitask missions autonomously.

And in order to do so, they need to speak to each other because each one of these agents are in their own multitask autonomous mission. If we don't secure these things, the, the, the threshold for chaos, right? The, the, the potential, you know, what could go wrong here is, is unlike anything we've dealt with, right?

I'm not gonna go Y 2K on you, but you know what I mean. This is, if we don't, this may be a case where we can't wait for the proverbial stuff to hit the fan before someone says, oh, we should really think about security here. We've gotta get in front.

Yeah, I couldn't agree with you more. And there's this whole body of work around, okay, what do agents need before we even think about trusting them to operate autonomously? Um, they're gonna need identities.

Just like people have identities. We're gonna need to be able to control access to other resources based on their as assigned identity. Um, they're gonna be able to start firing off 5G calls at some point, right?

So we've gotta watch how that filters down into not just my network, but you know, specific pieces of the network. Um, you know, does, does every, does every, uh, agent need to have its own, you know, virtual sim or eim, um, to watch it as, uh, we make calls. I, we agree that this is a huge body of work that needs to be addressed.

I also see attenuation, see what I did there with the 5G comment, um, or friction, um, against allowing autonomy to happen too quickly. Like, I think there's a huge process that vendors and system integrators need to go through to get enterprise IT shops comfortable with autonomous operation. Some are, you know, very few are more eager than others, but paying attention to what does it look like to develop trust in autonomous systems is gonna be a huge gate to adoption in just about any organization.

Scott, you're talking like a security guy. So let me ask this question because I'm sure someone in the comments is already thinking about it. Yes, there is this hesitance from the IT departments to move too fast, right?

Because we've seen what happens when someone rolls out an automated autonomous system that makes decisions. Either they don't like it because they feel like they're cut out of loop, or worse, it makes one mistake that then it gets amplified, right? Is the, is the speed aspect of it coming from the IT departments, or is it coming from the decision makers above that who said, well, I was told by Sam Altman that this thing will just run my, my company for me, and I don't need to think about it, but why is it taking so long for this to happen?

You guys promised that this was gonna be done already, while everybody else behind them is like, whoa, whoa, whoa, whoa. Wait a minute. We're not quite there there yet.

I've already laid off the people I was replacing with. Exactly. There you go.

Yeah, yeah. Well, so let, lemme give you an interesting factoid. I heard at a conference recently, I think it was 60% of CIOs are asking for bigger budgets for it, because they're being pressured by their boards to do more with ai.

Yep. And in order to do more with ai, they need to spend more on ai. And so they need to increase their budgets.

Uh, so Tom, that's your answer in my mind. I, I, I, I believe that that's real. I haven't seen that statistic, but that totally matches conversations I've been having.

Right? And if you think about budgets moving from actual personnel costs to paying for tokens for agents to run, like that's a real shift. I think we're gonna see over the next few budget cycles, um, that's directly follows that trend.

Um, I had something really important to say there, and I've, I've dropped it. Tom, do you wanna take it? Well, look, look, can I, I, I gotta talk, Tom, your point about people willing to trust autonomous.

So I had an interesting thing happen to me on Saturday morning. I had to go up to visit a, a relative in a hospital up about two hours from the house. And I, I took my wife's car, you don't usually drive it.

And I got on the highway 95, and this nice agent's voice came on and said, oh, we detect you're in a location, not a location or a whatever highway, and you could use the, uh, assist plus driving, just press the button on your steering wheel. I press the button on my steering wheel. Now, I've used cruise con control.

We've probably all used cruise cruise control before, right? It takes a little getting used to, but it's, this is not, this was full on driving my car. Sure.

So I will tell you that for the first 15 minutes, my heart rate was probably around 180. And, and I, I wasn't touching the steering wheel, but I had my hands right over because I'm waiting for something to go wrong, and I could go like this. And that took me maybe 10 or 15 minutes, and I, I kind of loosened my hands up a little bit.

My heart rate went down a little bit, wound up with, you know, my hands on my lap over here. But it was so g*****n strict, excuse my language, it was so g*****n stressful. Then I realized I'm better off shutting it off because I was too stressed out.

Hmm. You know, waiting for calamity to strike. I think there is that.

However, over time, I think people get used to it, and I think that's what we're gonna see here. Yeah, I agree. And to your point, I think think that there's, there's something you brought up there, like when people get used to the idea of it running, it disappears into the background, right?

Like, they know that occasionally they might have to touch the steering wheel or, or something like that. But there are also videos out there of what happens when the system fails. I think one of the ones was, uh, it actually was a, a major national news article where a self-driving vehicle would not stop at a railroad crossing because it's not something that it recognized and it didn't know that it was supposed to stop there.

Those are the kinds of exceptions to the rule that people focus on when they're talking about these things. Well, what happens if it uses the wrong OS on the upgrade? What happens if it does this?

And, and they're ignoring 95% of the things that work most of the time. Yeah. My calendar program might accidentally schedule me for a call at two in the morning instead of two in the afternoon once every nine months.

It's the rest of the time that it just works, that I'm just oblivious to. So is that part of the problem we're running into here is that people are just oblivious to the fact that we are securing these things, that we are trying our hardest to prevent them from becoming problematic. You know, something as simple, Alan, you brought up is like being able to steal data as quickly as possible rate limits, right?

Like, I can't pull more than a thousand requests from this particular device per minute. Sure. Not a problem for a human, not a problem for a well-written script.

A problem for somebody who's trying to dump as much data as possible before they get discovered and, and ejected. Like, are we, do we need to start thinking about security in a different way to make it seamless so that people just assume that everything works? Well, I so your comment, Tom brought me back to what I forgot a few minutes ago, and I'll just say this is all representative of a good and necessary tension between the innovators and the implementers, right?

And I think the innovation, um, around MCP and so many other things that we've seen has been awesome, and that's great. And we need people who are on the bleeding edge, hacking their way through the jungle and creating a path for others to follow eventually. Um, and that's the responsibility of the, the people who are procuring, implementing, and operating the tech to say, whoa, whoa, Whoa, whoa, whoa.

Let's make sure this is hardened for my use cases that I've kicked the tire sufficiently to make sure I can have a trust in it. And even if it's not full blown autonomous operation on day one, I've got something in place where I can start and I can map out a plan to say, what does building trust in this system look like? I, I think we're always gonna have those two, you know, things engaged in that tension.

On one hand, it kind of is what it is. On the other hand, I think it's good. And iron sharpens iron in this case.

Yeah, I would agree. Well, here, here's the other thing though, guys. And I, I, I've told this story before.

I'm doing a podcast for Network World about, I guess it's gotta be around 2010. So 15, 16 years ago, and I had the CEO of at the time of MongoDB, right? And Couchbase, two of the, at even then, two of the biggest, what we called no SQL databases of its time.

And I, and back then there was a lot of discussion, you know, that they, the security in these things was non-existent. And I remember asking them both, I said, gentlemen, a lot of people say no, sequel stands for no security. What's the deal with security in your products?

And they both agreed and said this out loud. I couldn't believe they said it out loud. We will put in more security when our customers demand more security and as much as security people, and Scott, I'm, I'm one too.

We sit here and think security's the most important thing in the world. How could you roll something out without security? If customers who are running and have established MCP as the, as the de facto standard are not saying, what about the security?

Right? Can't blame the, the producers, the developers for not doing more about security and that that's the, that's, you know, the world we live in. Well, it's, it's a fair point.

And, and I'll turn it into, you know, the, the last RFP that you read, right? There's a security requirement section in there, most likely, but it's not the first section. It's usually buried below.

Right? Um, because there is, you know, we're worried about that. Make the packets go make the product work first.

Is it gonna fit the business need that I need to drive in my enterprise? Um, and oh yeah, by the way, um, can you show me at secure, you know, the, it's not the primary thought. No.

I mean, this was a big thing. com in 20 14, 20 13, and then within a year or two, DevSecOps really showed its head and a there a lot of the DevOps people. Well, you don't need devs, you don't need SEC in the middle of DevOps.

DevOps has security built in. There's no such thing as DevSecOps. It's just a marketing term.

And I, I was one of them, I'll admit. Um, but I came to the realization that you did need DevSecOps. You did need to put security in the middle because you needed people, both security people and non-security people to recognize that security was important and it had to be part of the equation.

Yeah. And I, maybe we need M-C-S-P-A Context security protocol. Yeah, Yeah.

You know, something with security right in the middle, because that's what it's gonna take for us to make security, if not front and center, at least you know where it belongs. Well, I will, I will take that and like, I remain basically hopeful on all of this. And I would say, what if we could have this more unified view of what's in our sims, what's happening from a securities perspective?

What's happening from a network observability and streaming telemetry perspective? What if I could even pull stats on, you know, what's happening with my compute platform and different application performance? I think LLMs provide a unique opportunity to provide an integrated view of what's happening across those domains.

And I'd love to start a movement with you all right now to push our vendors to say, why not give me a unified view across my whole IT ops? You know, doesn't doesn't mean those silos go away. There is specific expertise that's needed in security, in networking, in application performance, but if I can create this shim layer or this umbrella that brings it all together, I think that could be really powerful.

Yes, we can. Tom, you were gonna say something? I agree, Scott.

Um, you're just gonna need to upgrade to the enterprise plus AI license for the tool. Um, that's gonna be eye wateringly expensive while I build it, and then everything's gonna be rosy. But I, nice, nice.

I believe you. I go back to something that, that I is kind of one of my favorite points to bring up here. And that's the idea that security that works well is not good because nobody knows that it's there.

Um, and all you have to do to figure that out is try to drive onto any military base in the world. Sure you have a government id, you are, you know, you are supposed to be there and they're still gonna check it. Uh, you know, think about, uh, the, the old RSA tokens, which are now apps on our phones or something as simple as a pass key, which is something I know that Alan loves, um, is there's still a popup, right?

Hey, confirm that you wanna log into this website using this pass key. That little tiny bit of friction proves that the security is there. You know, it's as simple as like having a little dialogue box popping up saying, we're au we're automatically logging you in with your credentials because you configured SSO you're awesome.

Like when we, when we let people know that, right? Like I, I, there's a story of IBM because I used to be an IBM or I was an intern there, and the road that went around the plant in Rochester, Minnesota was used as a shortcut by everybody in the, the community to get to Home Depot a little bit faster. And if you are familiar with any legal precedent in the United States, you know what an easement is, where if I have unfettered access to a road for more than one year, then I am, uh, then that is now a public road and I can't do anything about it.

So every year for one day a year, IBM would close, uh, the checkpoints at the front of the building, which were never otherwise closed in check IDs for people going through to make sure they were IBM employees. Uh, their stated goal was to, um, make sure that they didn't create an easement on that road. But the secondary goal was to remind people that there is a security aspect here, even if it's not alter very present now at the facility itself, you still had to use a badge to get in.

This was in 2001. It's like, you know, you had to swipe your badge and, you know, you had to, the doors were all locked and everything. Um, I will say that after I left and everything happened at the end of the year and nine 11, those checkpoints closed and everybody getting onto the facility had to check in for, I think it was two years before they would start allowing people to go through without having to show their id.

Like we, we have to make sure that people understand that there is security in the system, otherwise either A, they're not gonna care, or worse in my mind, they'll just assume that it's there. And then you'll, you'll see the breach notification, like, why were you storing all the tasks, the passwords in plain text in an Excel file? That's bad.

What deleting Now, Uh, gentlemen, any final thoughts on this topic? I I, I, I wish we had an hour and a half to talk about MCP and APIs and AI and everything, but, uh, I think this is definitely a fodder for another episode. Well, I would, I would say I am really interested in what AI tools can do to help us think about new security architectures, more holistic security architectures.

I think there's really interesting, um, promise there as well. So another, another follow up episode, Tom? Absolutely.

Alan, I, I would say this to, to wrap the bow around this one. Yes. In 11 short months, it's become the de facto standard.

As I said here today though, don't ask me if it's gonna be the standard next year, 'cause you don't know this stuff is happening so quickly. Yep. We don't know if something a better, a better mousetrap comes down the park, the road here and, and MCP, what?

Well, that's archaic, right? So let's not get, you know, we don't have to pull the emergency brake quite yet. Secondly, it has only been 11 months.

Security does, especially holistic, organically kind of embedded security takes time. We're not at the end of the game here. We're, we're, we're at the maybe the end of the beginning if that, and there's still plenty of time to get security right with MCP.

Well said, sir. Um, I, I wanna ask, you know, you two are both very busy, uh, folks, but I'm sure that you have a lot of great stuff going on. Uh, Alan, what are some cool things you've got coming up that people should be checking out?

Well, I am on Techstrong gang four or five days a week, uh, every morning from nine 30 till I guess 10 15. So, and we talk a lot about all this stuff there. I, um, I will be actually, I, I'm not sure when this podcast will, will actually air, but I'll be at the Qualys Rock, uh, uh, rock Con, uh, risk Operations conference this week.

I'll be at Cube Con in November, right? We'll be broadcasting live from there. And I mean, you could, you know, I I publish about 20 articles a week on the various techron sites, so you could follow me there, but I'm, I'm usually not hard to find.

Scott, what have you got coming up that people should pay attention to? So, for Network Automation Forum, we have our next event coming up in November the 17th through the 21st in Austin, Texas. Um, we're always looking for more automation of security content.

Um, so the agenda's closed for this one, but keep it in mind for, uh, for future events. There's a, that adjacency between the network and network security. I'm very personally interested in seeing how that goes together.

I will, um, see you all at the upcoming AI Field Day and the next networking field day in, uh, coming up in late October and then November. Um, and I'm always driving interesting conversations on network operations in IT operations, on the Total Network Operations podcast. So give it a listen.

Absolutely. And don't forget that for Tech Field Day, we have an exciting month going on. Uh, Steven is actually gonna be at, uh, tech, uh, tech Field Day Extra at Netex Insight this week.

Uh, next week we have Cloud Field Day Week after that we have AI Field Day, the week after that we have Networking Field Day. com and check out the calendar, uh, because we're gonna be busy and we definitely want you to take part in it too. Thank you very much for listening to this episode of the Security Boulevard podcast.

If you enjoyed this conversation, we would love it if you would subscribe on YouTube or in your favorite podcast application of choice. So you don't miss an episode. I'm a fan of Overcast, but whatever you want to use is great with us.

We'd also love it if you'd leave us a rating and a review, because that does help the show grow and help us find new audiences. com and the RUM Group. com, the Techstrong TV website.

Or if you are a fan of watching us on tv, use the Textron TV app. It's available on Apple tv, Roku, and any other smart device. Don't forget that we are also on social media.

Uh, you can follow us on or Twitter on LinkedIn. Just look for security BLVD for more content. Thank you very much for tuning in.

We'll see you all next week.