Machine-Scale Defense and the Future of Cybersecurity

Ann Johnson: [ Music ] Welcome to "Afternoon Cyber Tea" where we explore the intersection of innovation and cybersecurity. I'm your host, Ann Johnson. From the frontline to digital defense, to groundbreaking advancement shaping our digital future, we will bring you the latest insights, expert interviews, and captivating stories to stay one step ahead. [ Music ] Today's guest is someone who is redefining what it means to lead in the age of AI and digital transformation. Jeetu Patel is President and Chief Product Officer at Cisco, where he is driving innovation across security, collaboration, and AI-powered infrastructure. With a career that spans leadership roles at Box, EMC, and Doculabs, Jeetu brings a rare blend of product vision, operational rigor, and security insight. Welcome to "Afternoon Cyber Tea" Jeetu.

Jeetu Patel: Ann, it's so good to see you and so good to be on your podcast. What you didn't have on my bio is I am a 33-year-old friend of Microsoft. We have been working closely together at Microsoft in multiple capacities and a lot of different companies I worked at, it's a pleasure to be here.

Ann Johnson: > Ann Johnson: And I will say, we appreciate the partnership, obviously, our partnership with Cisco is special, but I want to give you a nod because you and I have worked together on a couple of projects and you're always pragmatic, and a gentleman, and we get to the right place for our mutual customers and that's really what it's all about.

Jeetu Patel: You're really too kind. It's a pleasure working with you as well. So, say hi to the team on my behalf.

Ann Johnson: So, at RSA this year which was just a couple months ago, you spoke about how AI presents both an incredible opportunity and a formidable challenge. Can you share with the community what those opportunities and what the challenges are, and how you view AI's role in the world today?

Jeetu Patel: Yeah, Ann you and I have-we've talked about this between our two firms as well, and I were to say what are the big, big opportunities with AI and in cybersecurity specifically, for the past 30 years, the one thing that's very clear cut in cybersecurity is that there is good people and there is evil people. The advisories are the bad folks; the defenders are the good folks. Like, there is no debate on that front. And then, when you start thinking about what needs to happen in AI, I think the scale in which these attacks are occurring right now, the surface area of attacks, is so-is broadening so quickly that the AI tools are being used by the attackers and the advisories and they're pretty skilled with it. The past 30 years, the advisories always had the advantage; they had the advantage because they had to be right once; the defender had to be right every single time. And so, the opportunity for us as defenders is to make sure that we can have machine-scale defense rather than, you know, human-scale defenses, because the attacks are machine scale. And so, that's the opportunity of making sure that we can use AI. And I think this one time, we are at this point in history where if we do it right, you might actually see the advantage flip to the defenders, because the way in which we build models, the kind of data that can be used to train the model which is all around defenses, could really help the scale tip in favor of the defender and that's what I'm hoping for. Now, the challenge that we have is, I think if you look at cybersecurity in general, the industry at large not Microsoft or Cisco or anyone, but the industry at large, the use of AI has trailed the use of AI in other industries. And it's because of a couple of reasons; the efficacy tends to be low with some of these kinds of initiatives on AI, and then the costs tend to be high. And I think both of us together as companies are working hard in changing this equation, but that's the thing that I think we have to do, is we-but that's-there is no opportunity to actually tip the scales in favor of the defenders and there is a challenge in that we are not using AI as dexterously as some of the other industries are and we ought to make sure that we are the critical infrastructure for predictable infrastructure; we got to do a better job on that front collectively as a community and, frankly, the software and the technology providers like yourselves and us. I'm really happy to see how the two of our companies engage and interact with each other. And we got to do that across the entire community, because the true enemy is not the competitor, it's actually the advisory. And we have to make sure that we do whatever we can to exchange data, cooperate, work well together, coordinate, build products together, integrate the products, even when it might not be in our commercial mutual interest, so that we can make sure that we can fend off the advisory effectively.

Ann Johnson: I want to pull the thread on a couple

Jeetu Patel: of things you said and I love that you're always so articulate and you're very crisp in how you talk about things, so it's helpful to the audience. I want to pull the thread on two things: One, you talk about machine-scale defense. We've talked in the industry historically about raising the cost to the attacker. And I think in this paradigm of the attacker using AI, we certainly don't have any chance

Ann Johnson: of actually raising the cost to the attacker or John Lambert at Microsoft talks about putting a lot of terrain between us and the attacker. So, can you talk a little bit more about how AI in defense gives us an opportunity to actually be that organizations that put more terrain between us and the attacker or raise the cost of the attacker, and then I want to talk about the ecosystem because it's a really important point. But let's just talk about what you're thinking about from an AI defense use cases.

Jeetu Patel: Yeah, I have two things, I think one of them is the experience for the defender and the second one is the technology that's underlying for AI which is the models that it used to go out and help us fend off the attacks. Let's talk about each one of those. I think for the longest time and, Ann, you and I share this passion, you've been in this industry for a while and so have I, you a little longer than me I think, because you were in cybersecurity before I was even, but I think what ends up happening is this industry is so insular in the terminology that we use that almost keeps people out, because we use so much jargon, so much complexity that actually becomes in intimidating industry for someone who is not completely entrenched in the industry to participate. And I actually feel like this is an area where having more people from the liberal arts background, more people from a design background, more people from a consumer and user experience background would really help this industry, because right now, everything seems complicated and we have to just make sure we simplify it. But I that both of us are working pretty hard on this for the products that we're building in each one of our organizations, but we need to continue to keep doing more on that front. But that's on one side, which is make it harder for the attacker to penetrate the system and make it easy for the defender to defend. Instead, we have it the other way around. We make it easy for the attacker to infiltrate and we make it hard for the defender to defend. So, we got to get that sorted out. And the second one is this notion around efficacy and cost. And I think if you look at the models right now, that are out there, I personally feel like we are entering into an era where the models are going to be for the underlying AI models that you have, will have to more focused on being more bespoke and more specialized to certain problems. There is a tremendous benefit to pre-training a model that is smaller in size with fewer tokens and actually do it in a specific domain and you will get higher efficacy than what you would get if you just did it in a very large parameter model with very large cyber tokens for training. And so, we did this at RSA where we launched and open-source model for security and frankly, you know, we took an 8-billion parameter model, pre-trained an 8-billion parameter model, that is now performing in our internal benchmarks and tests, as good as or better than a 70-billion parameter model. And we had taken 900-billion tokens and distilled it down to the 5-billion most relevant tokens that we trained the model on, and the consequence of that, was we were able to actually run that model on a single A100. And what that does is it gets your cost curve down precipitously as a result of doing that. And so, that notion of like let's make sure that we have bespoke models that are cost-efficient, that are high-efficacy, that can actually be generic models. The example I give tongue and cheek internally is, the model that you use to write poetry should not be the model that you use for national security and cyber defense. You might have to have a little bit more fine-tuning that is done on one of the models, and so that's an area where we are starting to see some early indication of success on efficacy and cost combined, going in opposite direction. That's at least the way that I see it right now, but those are the two big areas, or maybe three; experience, efficacy, and cost.

Ann Johnson: So, with that framing of experience, efficacy, and cost, you also talk about how innovation is becoming somewhat patchwork, right? Can you talk about what's driving that fragmentation; how we can actually somehow come together with the innovation too, because if we don't lower cost and we don't increase efficacy, we're never going to mainstream security tools on AI, because people-only the biggest and the baddest are going to be able to afford them, right? And then they're going to be worried about tuning the efficacy and they're going to be saying "Is it worth it?" So, a lot of that is how the innovation is happening and actually having innovation is more seamless than innovation is more streamlined. Before we get to ecosystem and I wanted to just mention that we're going to get there, because I think it's so incredibly important, but can you talk about how we can stop innovation from becoming fragmented and how we can bring it more back together and more cohesive?

Jeetu Patel: Yes, and that's a really important question. Right now, if you look at literally the past two or three decades in security, innovation was largely done as patchwork and what I mean by that is every single time there was a new threat, there was a new company literally that emerged that had a product that went out and solved that threat. And consequently, you have 3500 vendors in market and very few, you know, have any kind of concentration or share in the market and what's that done, is on average, people have between 50 and 70 products when you cybersecurity stack. And that's if you're lucky and if you're a large company you probably have something close to 150 and it's no longer tenable, because that's a 150 different management planes and that's a 100 different-50 different places where like contention can come between policies and you have all these kind of issues that are practical issues that don't allow us to operate quite the way that we need to. And I think this buying criteria, "Oh, let's just go do it an RFP" without looking at the experience of actually managing this entire estate of cybersecurity products that you might have; to improve your posture, has actually been not efficient in the market. So, I feel like there is a handful of companies, Microsoft is clearly one of them and you folks are doing a fantastic job, Cisco is clearly one of them where there are going to be these platforms that are formed. And I think there is going to be a handful of platforms that are going to work well together with each other. And there might be some overlap between the platforms. I would say that there is some overlap between the Microsoft security platform and Cisco's, but that doesn't stop us from working together. You can't have 75 products in your kind of cyber estate. To start with, you need to have a couple platforms, a couple management planes, a couple data planes and then the other products can fit in to those. I'm not all suggesting that we don't have a thriving startup ecosystem in cybersecurity; I'm suggesting that that should not create exacerbation of silos and fractioning of the estate when organizations are trying to manage it. And so, that's the thing that we have to actually get really sorted out well, and I feel like, you know, for sure that you will be one of the very prominent platforms in this ecosystem. Cisco is going to be one of the prominent ones. I would say there will probably be someone like Palo Alto that will be a prominent one. And we all just have work together and then just ensure that even if a company is not choosing head to toe from any one of us, we integrating with the rest and so that we come together. But more importantly, that we actually have a common management plane, all management planes and Copilots and all of those from an AI standpoint, all interface with one another. So, that's an area that, you know, I've always had this fantasy and where Microsoft's Copilot should have our skills and our Copilot it will learn which is, AI assistant should have some of your skills and we should just make sure that these are-these things are seamless and the agentic workflows just exchange data very seamlessly among our agents and your agents. And I think we are getting there. I would like to see the industry move faster, but I feel like Microsoft is one of the best partners that we have right now.

Ann Johnson: Thank you. And that brings us to this ecosystem conversation, security is a team sport. We have to be an ecosystem. And I don't want to just like hand-wave it and say clichés on that, "I agree with you." If we're going to truly build out the world of Agentic AI, we're going to have these agents that are skilled and doing specific things, so let's say your security operation center has an agent that's skilled out to do a specific type of hunting, they actually have to be skilled out to do that "hunting" across your platform, out platform, Palo's platform, CrowdStrike's platform whoever it is, or else the customer is not going to get that value; they're not going to be able to correlate the information. You're still going to have too much human capacity to do that and that's just one example. You know, and we talk a lot about having this agentic framework where the agents can take off the repetitive tasks, so the automated tasks and save the human brain power for things that are really complex and have the-we talk a lot and I think you're in line with this, about the humans become managers of agents, right? Well, there is no better place to do that than an ecosystem, because your human can become a manager of agents that can crawl across whoever's ecosystem and that's I think what the vision of, you know, as we-there's things we can do short-term, but as we're thinking about that Northstar and that vision, and really delivering things of value to our joint customers, it has to start with what the agents are able to do and the skills that they have.

Jeetu Patel: You bring up a really good point, which is, this move that we are experiencing right now from humans being tactical operators to strategic orchestrators, I think is a pretty important one. And right now, what ends up happening is we tend to be extremely fearful as a society on saying, "Oh, my goodness. This is going to take my job away." And I always tell people, I never worry about AI taking my job. I worry about someone using AI better than me taking my job, and I actually also worry about me not using AI effectively, delegating me in a position where I might not be able to do my job. So, those are things that we have to kind of think through, because I think you, you hit the nail right on the head which is, our agents in our in Cisco should actually work with your agents, exchange data or make sure that there is identity verified across those agents. There might be times that those two agents disagree with each other and they have to come to some kind of a resolution. There might be an orchestrator on top that's actually even an agentic orchestrator that first goes out resolves some of the differences and then you might have a human in the loop that's actually made me show that they can approve the final recommendation. And there is a human on the loop that's actually observing what's going on. I do feel like those are aspects of the agentic world that are going to become very, very exciting and Andrej Kaparthy had a very interesting Tweet on products that are built are going to need to be built not just for humans, but also for agents, because any product that's built that is only for a human and not for an agent, is not a text-first interface, has too many proprietary ways of storing the information, has too much clutter from a visual UI perspective. Might not necessarily be the product of the future, because you will not be able to have agents engage and kind of Copilot the opportunity along with an assistant. But I think that's a very important point of how these products are going to need to evolve over time.

Ann Johnson: We're going to have a proliferation of agents pretty quickly.

Jeetu Patel: Yeah.

Ann Johnson: So, the ability to orchestrate them, the observability of them, the explainability of them, and even the identity provisioning, deprovisioning, least privileged, there's so much to contemplate. If we as an industry can come together on what those standards are, we're ahead of where we ever were on a lot of different things, right? If we can align and say, "Yeah, here is the standards cross agents, here's what we want for explainability, for observability, identity provisioning, deprovisioning, etcetera," we're going to help our customers so much, because when they go to deploy the agents they don't have to worry about "Oh, this is an agent from Cisco. It's going to-it needs to be provisioned a different way blah-blah-blah." It's going to, you know, so completely agree.

Jeetu Patel: We think like a REST API or.

Ann Johnson: Yeah.

Jeetu Patel: Agent communication, which is not an API, but which is an easier way to have it for recall the comunica-I think, by the way, Model Context Protocol or MCP was.

Ann Johnson: Yeah.

Jeetu Patel: A fantastic addition. I think agent-to-agent, and I was really happy to see Microsoft kind of lean into that, but those are all kind of steps forward where you're now starting to see-and I was really happy to see OpenAI adopt MCP, which came out of Anthropic. We just need to have that happen so that more and more people adopt the standards and then just move forward and everyone invest in progressive standard rather than each one of us fighting on a proprietary standard which doesn't help anyone, you know?

Ann Johnson: Correct. And I think it's a good way for us to press reset, right? The industry is going through a huge transformation. Why not reset now and do it right from the start?

Jeetu Patel: Exactly.

Ann Johnson: So, let's talk a little bit about AI, AI as a whole, right? AI is being used across a lot of sectors. It has applications in health care; it has applications in just about anything you can think about. What do you think are some of the biggest risks if AI is deployed without proper guardrails for cybersecurity, like if the AI, you know, folks were innovating and they're loving what they're doing, and they're seeing so much progress with AI, but they're not engaging the security team or the security team isn't providing the right guardrails. Where do you think we end up?

Jeetu Patel: That's a really important question and one where, you know, we've been thinking about this quite a bit, and one of the consequences of cybersecurity and safety not being handled effectively with AI; what happens if the systems of AI are not trusted by the user? Like, what's the consequence of that? In the past, past few decades, security was always seen as something that was at odds with productivity. You always have this kind of tradeoff that people talked about making which is do you want to be secure or do you want to be productive? And, you know, it's a matter of how much risk you take on the security side in order to be productive, whereas now what you're starting to see is security is becoming a prerequisite for AI adoption and, therefore, for productivity in the next level. And if you don't have the right level of safety and security in place and you don't have the right level of trust established with the user, you cannot accelerate, you know, adoption. And that to me is actually, in some ways, exciting as a time for the security industry, because what's happened is in some ways it was a failure, in some ways it's a success. The reason it was a failure is because people are now seeing what the profound level of damage can be caused as a result of a cybersecurity breach. Like, lives can be lost, right? Hospitals stop working, people can't get dialysis because there is a cybersecurity breach. The water supply stops working at your water plant, the power grid stops working; these are like meaningful critical infrastructure elements that can cost lives for cities, countries, nations, continents, so on and so forth. On the other hand, I do feel like this is a very interesting time, because there are 2-dimensions of cybersecurity; one is have to use AI to have machine-scale defenses and that's super important. And then the second area is you have to secure AI itself. You have to secure the models, you have to secure the agents, you have to secure the applications, you have to secure the clouds, and you have to have a level of neutrality where it works across all of them. That, in my mind, is where the opportunity lies and also where the risk lies if it's not done well. I'm actually, right now, not positive on the progress that's being made, but I'm also a little paranoid that we're not moving fast enough as a society.

Ann Johnson: Yeah, and I think there is so many and we could spend the rest of the day talking about the use cases, you know, things I think about for AI that are for the greater good, like predictable water supplies, clean water supplies, predictable food supplies; where is the world's climate changing going to have the most impact on population so we can actually predict orderly migration patterns, right?

Jeetu Patel: Yes.

Ann Johnson: All of those things, there's just so many things I think AI can do for the greater good and not just for security. But we have to make sure the data that is in all of those things is really sensitive data and that's what we have to protect.

Jeetu Patel: One other thing to keep in mind is, not all AI competitors and all countries will operate from the same set of rules.

Ann Johnson: Correct.

Jeetu Patel: And you will have to make sure that that's also being thought through very carefully, because privacy might not be a core value for every country, but it is for us and we have to make sure that we honor it, but we have to make sure that we figure out a way that through innovation, the privacy is honored, but speed is not sacrificed. Those are kind of important dimensions to kind of keep in mind and think through as we are going through this. And then the last thing I would say is, very personal, but right now, I think what I'm most excited about of anything with AI is the impact this can have on the health care industry, and it is just phenomenal what the upside could be on diagnoses of diseases, of prevention of diseases, you know, like I currently have both my in-laws in the ICU right now. And both of them, ironically at literally the same time, they're one floor apart from each other being treated by the same surgeon; they're going to have-they're facing some serious kind of issues on their limbs, they face amputation. And these are all areas which can be prevented with normal health of humanity in general can get better where we just have less suffering on the planet as a result of it, and I'm sure excited about that dimension of what AI can do for the world.

Ann Johnson: I think so also, and I didn't even mention health care, but to your point, it's the ability to modernize, to truly modernize medicine and make it really targeted and proactive and predictive, right?

Jeetu Patel: Yeah.

Ann Johnson: Defeating cancer before it ever starts.

Jeetu Patel: Exactly. Exactly, yeah.

Ann Johnson: Well, I want to ask you two last questions. One.

Jeetu Patel: Yes.

Ann Johnson: We've talked a lot about, you know, talent shortages in cyber, but the world is changing. Someone actually asked me at the RSA Conference if I were-if I had to make a choice on the hiring of a cyber-person, AI person, what would I hire? I said, well that's-that question is a bit of misnomer. And I said like, we still need to skill-up cyber people, where we need to skill them up with AI skills,

Jeetu Patel: so you're not talking about hiring a different person,

Ann Johnson: you're talking about making sure they have the right skillset for the future, right? How do you think about it-how-you have a lot of employees, you're an 8000-person company, how do you think about getting your employees skilled up for this next generation?

Jeetu Patel: And the way I think about this is no one in the world is sufficiently skilled in AI and no one in the world can afford to not be sufficiently skilled in AI moving forward. So, we all have a lot to learn, but there is going to be no cyber person that is not an AI person moving forward, so I completely agree with you. I think it's kind of a false choice. You have to make sure that you're dexterous in AI or you're not going to be relevant and you're going to have to make sure that, especially if you're in cyber that the dexterity in AI is going to be the difference between success and failure. I would never want to hire a person in cyber that is not curious about AI and is not committed to learning about AI. I don't care if they're not the expert today, but I want to make sure that they have the commitment, because actually what I've found is the time of getting from novice to expert, has compressed so much because of AI. That, if you have the hunger and if you have the curiosity, the path to expertise and mastery is actually not far. It's just a matter of will, persistence, and desire to get there. I would not hire anyone who is not curious about AI or is debating AI at this point in time. I've publically told this to people at Cisco, if you don't believe in AI, you should probably not be at Cisco, because you would be the wrong fit.

Ann Johnson: Yeah, and I think that's fair and I think that that's a legitimate thing to say now, because we're moving so fast, right? And I've been in tech forever and cyber 25 years, as you had mentioned, and one of the reasons I love it is because the industry change and you have to skill yourself up regularly and change regularly. I find it refreshing.

Jeetu Patel: And it kind of got boring for a while, so it's exciting again, you know, like there is new ways that you can go beat the bad folks. > Ann Johnson: Yeah. So, we close "Afternoon Cyber Tea" with optimism. With that in mind, considering everything we also talked about today, what are you optimistic about for the future of cybersecurity? I'm optimistic about the fact that the defender might be able to have the scales tipped in their favor and get the upper hand where we don't no longer have to have the advisory always be the one with the advantage because they have to the right ones and defender had to be right every single time, but because of the way that we can actually create cyber defenses on massive scale, you might see the advantage tip in favor of the defender and if that happens, that's going to be a massive step forward for the progress for humanity. So, that's what I'm most excited about. It unlocks the potential of AI in such a profound way that the ripple effect of success in cyber, is humanity having the ability to live their full potential because of AI, and that to me is a whole different level of mission and purpose to be in this industry for.

Ann Johnson: I love that. Not everything we do, as you know, is always altruistic. We work for profit organizations, but.

Jeetu Patel: Yes.

Ann Johnson: We do try to make the world a better place. One of the reasons I love working with you, is know that you have that ethos. So, I want to thank you Jeetu for being on the show. I think it's been fantastic and know how busy you are.

Jeetu Patel: Ann, it's such a pleasure to partner with you to be on the show. Thank you for having me and I can tell you what a big difference you personally have made in the relationship that we have and I'm so grateful to you for the friendship and for the partnership that we've enjoyed, and I am hoping that we are just getting started.

Ann Johnson: And many thanks for our audience for tuning in. Join us next time, on "Afternoon Cyber Tea." [ Music ] I invited you to join me, because AI continues to be top-of-mind for security leaders and practitioners. He really understands the questions security teams are asking as they adopt AI, because he's asked those same questions throughout the entirety of his career. Jeetu also offers real solutions to recommendations and his insights will be beneficial to anyone in security regardless seeing the world. [ Music ]