AI Innovations in Cybersecurity with Anand Oswal | RSAC Conference 2025

Hey, it's Techstrong TV coming to you live day three of our coverage of RSAC here at Moscon West in San Francisco. We've had great conversations with leading cybersecurity firms and experts from across the globe, and I'm pleased to welcome back one of our guests, Anan Sal, who I've had the privilege of interviewing before the SVP and GMF network security at Palo Alto Networks. Anan.

Great to have you. Big week for Palo Alto. Great to be here today.

Yes, Lisa. It's a great week. Can't Go to a conference these days without talking about AI and security.

It's a stable stakes. Yeah. But Palo Alto has been really obviously making headlines, especially this week, announcing the acquisition of protect ai, new AI powered security platforms, because every company has to have an AI story.

Yeah. Talk to us about the impact Yeah. That protect AI will enable Palo Alto to make with its customer base.

Yeah. So if you step back, Lisa, you talked about, uh, you cannot talk about ai. Uh, you know, a year ago, a, we talked about assessment shift that AI was having on businesses.

Yeah. Right. At Palo Alto Networks, we solved two of our most customers more important problems.

How do you have employees access AI applications safely and securely? And as you're building these applications where every organization is building them Yes. They wanna change their business, they wanna give new experiences to their customer.

How do I deploy these securely so that in mind we've done, we've done two things. The first is for employees. Look, look, everybody's accessing AI applications.

You are. I am. What employees are because they want to get more productive.

Yeah. They wanna be more effective. They wanna be more efficient.

Now, all of this is getting access from the browser. The browser is the prominent attack vector. Mm.

So a new secure browser is needed for this new era of ai. And what we launched today is that a secure, we already have a secure browser. We said employees can browse bravely, they can access these applications not worrying about, um, data leaking, et cetera, because we are making sure that the organizations, all the tools for them to do it.

Advanced threats, advanced malware, which are browser native. But the most important thing also for, for users is that don't compromise my user experience. Right.

So we will enter their web and SaaS applications, give maximum performance, reduce your reliance on all this legacy VDI stuff. Yeah. At the same time, have that consistent security.

That's what we did for employees accessing AI applications. The second thing I think you do touched on protect. Yeah.

And, and what we launched this, uh, this week is Prisma airs is the industry's most complete and most comprehensive AI security platform. Look, we look at, um, AI application changing the landscape. Yeah.

Right? You have your app architecture's evolving. Yeah.

You have new types of threats. You can't have point products for different parts of your thing. No.

Right? No. It all the stitched together delivered comprehensively and solving the entire problem for the customers.

Yeah. Well, doesn't the average organization have seven to 10 plus different security tools in their environment anyway, More than that point solutions average. The average, the average customer has many more than that.

But if you look at ai, it is changing the way you build applications. Right? Now, you and I been around the block a long time.

Yes. Long, long ago. How applications built a three year model.

You had a front end, you're a database and you're a backend on the application. Yeah. Then key along the cloud, it modernize your application.

Microservices, cloud constructs, AI applications are really the third wave. You're bringing in newer things. It's not just an application, a model, and you're done.

Yeah. You're bringing in infrastructure models, data tools, plugins. What happens, it gives you attack surface.

And the complexity. And the complexity. Yes.

So you can fight those with your point products for model scanning, point products for posture point products for red teaming point products for runtime point products for our agent security. You gotta solve this cohesively and consistently across the entire gamut. And that's what we announced with PRIs Myers.

That's fantastic because you talked about the attack surface. It's just gonna continue to spread. And I always think of it as it's really amorphous.

There's so many new applications, threat factors, channels, actors, that this problem isn't going away. Yeah. It's just going to continue at speed and scale.

Yeah. I think we call it the scale, the sophistication and the speed, the three S's. Yes.

Yes. And AI is just double charging all of this. Yes.

If you think about AI applications, they're bringing in new threats. Yeah. Attackers are, uh, using prompt injection techniques to get customer data.

So they, they pretend I'm you and they'll get my data. They're doing, they're making code generate malware. You can do a model dos attack.

You can do a simple query to a model, like a simple example. We print hello a trillion times. It can make the model spin.

Of course. Those are simple ones that people can block now. Yeah.

But most important, you don't want your sensitive data. You've trained yourself, you've taken your data, you've trained your models. You want that data to leak.

Right. Now the other thing that's interesting, I know you said about ai, the next most famous word is agents. Right.

Especially these days. Every Everybody saw. Yes.

So if you think of agents, what do the agents do? If LLMs give you answers? Agents give you action.

Action. Right. They plan, they adapt, they act autonomously.

Yes. And they replan and they react. Yes.

So now when you're thinking of security, you need to think how the model is thinking, how the model is behaving, what the model architecture, the risks are no longer in just your core and vulnerabilities. And then the training data user train the models. And that's why when we announced PRI Myers, we said there are five important pillars, okay.

Of AI and time security. First model scanning. You need, we, we scan code and infrastructure today.

Yeah. So what's different in models? The, the difference is the data user train the model, different model architecture, different model behavior.

Then we think of posture. Posture should not be just for model. Again, like I said, we don't need point products, no network application models, data agents, all this done.

Comprehensive posture management. Then red teaming. What does red teaming do?

Red teaming is trying to mimic how things, so you think of AI and agents, we don't execute code only. We plan, we adapt, we replan. So your AI red team should be autonomous.

You should be able to think how and adverse thinks and be comprehensive to mimic real behaviors. Okay. If you think of a runtime security, you have threats that we know from classical applications, classical threats.

Then you have specific threats for AI application. The one I talked about, prompt injection, model dos, et cetera. But then you have other threats related to agents.

One important one is, look, agents will give you answers short term, but agents wanna be personalized for you or the long term, whatever poison the memory of that agent uses, now you're gonna, it's gonna change the behavior of the agent. Okay. Agents have to do autonomous actions.

What do they have? Excessive permissions. Yeah.

So all this needs to be thought through comprehensively to make sure that you have security. And that's what Prisma airs is the most comprehensive platform. Discover your system, assess your risk, and protect your, all your threats in one platform.

Single management workflow is completely, uh, unified. So it's not dealing with 12 Yeah. Different point products to solve the same problem.

Well, the, what I'm hearing is massive simplification. You have to, for CISOs, because the landscape, the threat landscape will continue to evolve. The technology will continue to evolve.

Yeah. There's no slowing down. Nobody wants less data.

Slower. We know that. Yes.

But something else that you said that I like, because I always think of humans in the cybersecurity chain are often the weakest link, but they can be the biggest asset. So what you're enabling brave browsing, you're enabling humans to take some of that risk out of the equation. Yeah.

Which is essential because employees need to be able to deliver what the customers want. What do the employees want? Employees wanna be productive.

Yes. They wanna look good. They wanna get their work done effect, you Wanna look good And they wanna get it done efficiently.

Yes. So they wanna use these tools. The job of the organization is to ensure that, hey, do you have safe and compliant usage of these tools?

Yeah. Can I control what applications are being used and what I don't want to be used? Can I protect the sensitive data from leaking out the organization?

And all these applications, Lisa, they give responses back, but the responses have threats and malware. Right. Right.

Protect you. That's what you wanna make the, uh, empower the employee to do. Yes.

And that's why we say browse bravely. Love that. Now, if you think about developers, they're building applications because they'll transform your business.

AI applications are gonna change every business. Absolutely. You're gonna change your, uh, your p and l is gonna change the way you think about customer experiences.

You're gonna gonna give new experiences. Now you wanna deploy those applications bravely, but a lot of times if you don't know all the steps that you take to make sure it's done right. Right.

You could make a mistake. Absolutely. 6 million models on hugging face.

6 million. Yes. Wow.

Now, a developer could just download a model, but what if it has vulnerabilities in it? It has malicious code inside it. Right.

So you just scan it. What, what if your agents have excessive permissions because they are acting, they're acting autonomously. Right.

And they have excessive permissions. And they, and when you do that, something else happens. Yeah.

So all that needs to be thought through model scanning, a posture management, red teaming your runtime security. And we, the platform should be ready for newer things like agents. Right.

But that's coming. The plethora of agents gonna happen. It's Coming.

There's no slowing that train down at all. I, I talk with a lot of CMOs and even in the marketing function, CMOs, part of their KPIs is AI agents acting on their behalf. Yes.

It's it's, everyone's embracing it. It's kinda like, well, I I always feel like when chat GPT was born a couple years ago, this catalyst just went haywire. 'cause AI's been around for a long time.

Yes. But suddenly now it's, everyone has to have an AI story. Yes.

Well, there's a lot of opportunity. Oh, yeah. But there's a lot of risk as well.

Yeah. How is the CSO role in your experience changing to be able to have this AI platform comprehensive view Yeah. And also enable those employees Yeah.

To browse the brain. Yeah. Brave of brows.

If, if You look at the organizations, the, the job of the, the security organizations is to of course secure the organization. Yeah. You wanna make sure that the employees, that the sensitive data is not leaked and in many cases is not to malicious intent.

Yeah. Employees may not know you put some things in your AI tool and the data is out. Yeah.

You didn't do it intentionally. No. So how do you have the right controls?

Full visibility. It starts with full visibility. You can only secure something if you see it.

Right. Right. Can't secure what you can't See.

You can't secure what you can't see. So full visibility, then you can decide do you wanna allow it or deny it, or you wanna block or limit usage. Right.

Once you do that, then what? For the applications that allow, how do I ensure the right policies that my sensitive data is not leaking out? That's what you wanna do for, uh, for employees to be productive.

Yes. But at the same time, do it safely and securely. Right.

Right. And the same thing applies to the, the way we approach you building applications for your end customers. Right.

If you're able to use all these tools that make it more efficient, you're able to give new experiences to customers potentially generating new revenue streams. Absolutely. Reducing your cost.

You wanna do it securely and safely. And That's business value. That's outcomes to the business.

Exactly. And that's what the, the C-suite wants to achieve. Everybody wants that.

Yes. So, so you just need to make sure that you're thinking through all aspects of security. And it's not an afterthought.

Can't be. I think that's been proven time. But again, because the, the sophistication of the risks and the attack, they're growing, the technology innovation is growing.

Is it possible to fight fire with fire? Are we, are we gonna be able to be proactive here? Yeah, Absolutely.

Yes. The answer is absolutely. That's a good answer.

I'm happy to hear it. Let me give you some data points. Today we are blocking on our platform, 31 billion attacks every single day.

31 billion attacks every single day with a B? With a B. Yes.

Wow. Right now, a small number, 9 million or so on net new attacks, day zero Attacks every day, attacks That nobody has ever seen before. 9 million new a day.

Yes. And the reason we are able to do it is because we use what we call precision AI security services. Okay.

It's a combination of machine learning, deep learning, and all the variability we can get through gen ai. Right. Because the days of you getting infected with something, me learning about it, building a signature, patching my system so everybody else is safe mm-hmm.

Are over. Right. I wanna protect things that you have never seen before.

Right. We have over 4,400 deep learning models on a platform that are able to look at, um, content, metadata, real time traffic to stop these Yeah. Right there.

And that's the power of what we can get through AI to solve these things holistically. So is that where CISOs need to be focusing on all of the data, the, the, the net new attacks be able to get ahead? It's both.

Right? It's both. So of course you wanna stop all, all the attacks that are, that you know of because it's easier now to create attacks with ai.

Yeah. But you also wanna have a system that is able to, uh, to look at all your data, to look at the variability, to look at your behaviors and stop threats that you've not seen before. You need to Be doing both simultaneously Because at the end of the day, it, it's not enough to say that you got infected.

Everybody else is now secure. Yeah. Like, I don't want anybody infected.

I wanna be able to be proactive Yeah. Right now. Yes.

And that's what we've been working on. Yes. And that's where we wanna make sure that we are ahead of the game.

That proactivity is so critical because of the speed with which everything is, is scaling the good stuff, the bad stuff, the questionable stuff, the opportunities. What's been the customer and partner feedback this week since the announcement of the acquisition. I've, I've probably met 30 customers also in the, already this week.

Yes. Already. Wow.

On busy week. Someone on some together. Okay.

It's just, it's just, uh, it's a, there are a couple of things that have come out. Every customer saying the point you said before, I have too many tools. Yep.

Too many point products. I don't know how to make this work. Mm-hmm.

Help me. Right. Uh, I, I can't have a consistent policy across all my infrastructure.

This is just becoming to a point where I am not, I need help. And they're overwhelmed. I'm sure.

Yes. Oh, they're overwhelmed. Yes.

The second thing they're saying is that AI is a good enabler, but how do I make sure I enable it and, and stay safe and secure both for the employees and the business value when I'm building applications. Yes. And, and the third is that, like, what does it mean for me in terms of how do I ensure that all of these things come together?

Yes. I reduce my operational cost. I am getting more and more efficient, and I'm able to stay at the curve or in a simple way of saying, how, how does my organization make more money?

How do I save money? And how do I bureau outta trouble? So are you seeing more of the CISOs now having to go to the C-suite, to the CEO and, and prove business value and AI spend?

I imagine they do. It's no longer just a, yeah. A lot of the AI spend on building new applications, it's coming from the business.

A lot of the things that you need to do for, for preventing, for security, for, for secure, for, uh, securing usage of AI applications is definitely one security group. So it's a combination. Yeah.

Yeah. And it varies by organization. Different organizations are structured differently.

Right. So Walk me through some of the plans for existing customers from a migration integration perspective. What can they expect to be able to really capitalize on all of the value that Yeah.

That Prisma Airs is gonna deliver? I think that's a great question. So if you think about our, our network security platform, right?

It is comprehensive. It, it, the whole, the whole idea is any user on any device accessing any application, any data on any network consistently secured. So when you have new use cases like ai, the platform is extensible.

So you take the example of employees accessing AI applications, it's easily enabled on the platform no matter where you are. You could be in the office going through a firewall. You could be home going to SSE if you think about your developers as they build new applications for the business value we talked about.

Yeah. The existing platform is extensible for me to enable all the capability I talked to you about with Prisma, with the same framework. So now you're leveraging what you have to provide a consistent capability to the customer.

That consistency Is, it's not yet another point. Product, a new ui, a new tool, a different policy for ai. Because I mean, if you fast forward a few, maybe, I dunno how long it is hard to predict.

Every application will be an AI application. Yeah. There's no distinction anymore And it's not gonna take too long.

Yes. So you won't have that extensibility of the platform because you'll always have new things, or you look at a secure browser, it plugs in exactly. Into our sass e architecture.

So it's not like another solution for something. It's all integrated and bought together very cohesively for our customers. So easy for them to consume, Easy for them to consume, which is great.

What are they looking at timeframe wise to be able to really extract the business value here? Yes. And, and also dial down that technical debt of all those extraneous security tools.

Yeah. So I think it, it varies. Every, every customer is in a different journey.

Sure. We have many customers today who are using the complete platform, securing how they have application in the data center, securing their cloud assets, securing their ai, both for employees and applications and the remote workforce. But the platform's modular, you have a, you have a choice to start in a different way, in different journey.

Yeah. So you could start with your, your SASS e customers where your remote workers and remote branches, and then migrate to other parts. Um, majority of our customers, of course, are using our firewalls hardware and software and the cloud and the data centers to protect, and then they're able to move there.

So it varies on customers, Many pathways of opportunity, Many pathways to get there. But the end goal is the same. How do I get a consistent security?

How do I reduce the operational costs? How do I get into the a better ROI for my investment? And I just wanna make sure this security works.

I can't be reactive, I wanna be proactive That you can't afford to be pro Yes. Reactive anymore. We have to be proactive.

Yes. But it's, it's a balance. Yeah.

But it's also about, if you see the majority of, of, of, um, issues happen because of manual configurations of this configuration. Yes. Yes.

So what we are doing in the platform is easier way for you dynamic determine. So the example I gave you on red teaming. Yes.

After I do my red teaming, the policy recommendations are based on two things, my best practices and the environment. Yeah. And that's tuned dynamically.

So now you don't need to do the hard work of figuring out what's the right policy. I'm able to recommend that to you and just single click apply it. So workflows are getting more streamlined, More streamlined, more simplified.

Yeah. Right. So get, I get the value to the business, the value to the, to the cso, the value to the C-suite is this, I always kind of look for what's the bridge between the developers and the security folks?

Because we know DevSecOps as a concept is, is still kind of in its infancy. Yeah. Is this a facilitator?

Is this that platform, that bridge between the developers having the experience they expect Yeah. And the security folks being able to ensure the security of the environment. Yes.

I think it's a very good question. If you think about the journey of cloud, the developers went there before the security people came in. Yeah.

So what are the first question the security professionals ask? What's running in my cloud? Right?

Is it exposed? Does it have vulnerabilities? Once you give them the list, it's like, it's too long.

They shorten that list for me. Yeah. I, I can't deal with all of them.

It's overwhelming, Right? Yeah. Then they say, look, I need to have runtime.

The idea is that every, all these pieces, and the same thing is happening with ai. All of these pieces need to bot with the cohesively. So you are providing a unified solution, not a piecemeal solution of, Hey, this is my vulnerability, this is my posture.

This is the results of your red teaming. This is your runtime risk. No, tell me all of them.

Show me a workflow to link them all to gimme the best outcome. Don't show me all this small activities. I'm outcome based.

That's what they want. You Wanna get rid of that noise Exactly. Through the noise.

Yeah. To elevate the impact. Yeah.

And, and in the end, give and show the outcome. What is happening. Otherwise, I'm dealing with 10 different products, 10 different solutions.

They have 10 different management planes. They don't talk to each other. They don't check their intelligence.

I'm not getting the outcomes I want. Right. What excites you about where we are in security in 2025.

Here we are with about 45,000 security professionals and vendors and partners. What's, you mentioned some good news earlier, but we're gonna be able to get proactive. Yeah.

Palo Alto is enabling organizations Yeah. Across industries to get there, which is table stakes these days. But what excites you about where we are from a, an offensive perspective in cybersecurity?

Are we there yet? Yeah. Look, I think the, the more important factor that I'm excited about is that I think for the first time with, with the advent of ai, you feel that security is solvable.

You can really make sure that you can stitch all these things together cohesively to solve customer problems. But you have to have the right, right approach. Sure.

If you go with the 10 different point products for 10 different point solutions that are not, not talking to each other. No. Not sharing the, it's very hard.

Yeah. Right. It's very complicated.

So how do you have that consistent policies? I, I could be at home, I could be in the office, I could be on the road. I could be on this device.

I could be on my personal device that's owned by me. I could be accessing an application to my data center, cloud, SaaS, ai, it Doesn't matter. It shouldn't matter This.

And that is exactly what we're solving with the platform. Yeah. Awesome.

What's next for Palo Paloalto? Obviously great momentum this week and gonna continue that. What can we, any nuggets you can share with us?

Always be new innovation that you'll see from us in cybersecurity. Cybersecurity, ever evolving field. Yeah.

Stay tuned for more information. I love that. Anna, it's been great having you on text, on tv.

Thank you for sharing and really dissecting what's new at Palo Alto, how you're enabling this comprehensive, cohesive view. You're taking out technical debt, you're simplifying the CISO's workflow. You're simplifying the employee experience in this age of ai that is so incredibly important.

And you said, in the age of ai, security is solvable. I love that. Thank you for all your insights.

I appreciate it. We'll be following Paloalto. Yeah.

Thank you Lisa. Great to have you. For Anand Oswald, I am Lisa Martin.

You're watching Text on TV Live from day three of our coverage of RA See. Stick around. We have more great content coming at you on text, on tv.

You will see you in a minute.