When the Ground Keeps Moving: The New Reality of AI Security

By the time this episode reaches your feed, the AI landscape may have changed again.

That's becoming the new normal.

It's tempting to chase every new model, every policy update, every headline. But the organizations making real progress aren't rebuilding their strategy every time the landscape shifts. They're building the resilience to keep moving anyway.

Madison Horn and Rob Geis join AI Proving Ground Podcast to explore what this new reality means for enterprise security. The conversation moves beyond the latest AI models to something much bigger: how to build an organization that can absorb change, contain risk, and stay effective when the technology refuses to stand still.

In AI security, keeping your head on a swivel isn't just good advice.

It's becoming the job.

Support for this episode provided by: Cloudflare

More about this week's guests:

Madison Horn is National Security & Critical Infrastructure Advisor at WWT, where she helps organizations strengthen cyber resilience and prepare for emerging threats at the intersection of AI, cybersecurity, and critical infrastructure. With more than 15 years of experience spanning national security, global incident response, and cybersecurity strategy, Madison advises enterprise and public sector leaders on securing the technologies shaping the future.

Madison's top pick: The Quantum Signal from the White House Operators Should Not Miss

Robert Geis is a Field CTO for Global Cyber at WWT with more than 30 years of experience helping organizations strengthen cybersecurity, manage risk, and protect critical infrastructure. He advises enterprise leaders on cyber resilience, governance, and AI adoption, helping some of the world's largest organizations navigate an increasingly complex and fast-changing threat landscape.

Robert's top pick: AI Security Threats: Mythos, Project Glass Wing, and the New Cyber Resilience Framework

The AI Proving Ground Podcast leverages the deep AI technical and business expertise from within World Wide Technology's one-of-a-kind AI Proving Ground, which provides unrivaled access to the world's leading AI technologies. This unique lab environment accelerates your ability to learn about, test, train and implement AI solutions. 

Learn more about WWT's AI Proving Ground.

The AI Proving Ground is a composable lab environment that features the latest high-performance infrastructure and reference architectures from the world's leading AI companies, such as NVIDIA, Cisco, Dell, F5, AMD, Intel and others.

Developed within our Advanced Technology Center (ATC), this one-of-a-kind lab environment empowers IT teams to evaluate and test AI infrastructure, software and solutions for efficacy, scalability and flexibility — all under one roof. The AI Proving Ground provides visibility into data flows across the entire development pipeline, enabling more informed decision-making while safeguarding production environments. 

Transcript

Keep Your Head On A Swivel

SPEAKER_01 0:00

Growing up playing sports, I heard coaches repeat the same phrase whenever the game started moving faster. Keep your head on a swivel. And what they meant was don't become so focused on the person directly in front of you that you missed what's developing somewhere else. That feels like the right mindset for enterprise leaders navigating frontier AI today. Mythos, Fable, GPT-5. At the pace things are moving, anything we say about the status of these models could change by the time you're listening to this. Access was granted, revoked, and partially restored, all in a matter of weeks. It's all moving at breakneck speed. But that volatility is the point. Because when a CIO asks, what should we do about Mythos? They're rarely asking about the model itself. They're really asking a set of key questions like, how can I defend at machine speed? What tools can I actually trust? And who's accountable when a dependency changes or disappears altogether? These are questions about how the enterprise operates amid an AI landscape that can, and almost always will, change when you least expect it. So in this episode, we're talking with Madison Horn and Rob Geist from WWT about building a security architecture and operating model that can withstand that constant change and what it'll take to absorb an attack, contain the damage, and recover at machine speed. Madison is chief advisor for national security and critical infrastructure, and she brings a profile that is genuinely unusual in enterprise technology, a cybersecurity exec who has led red team operations, built global cyber defense practices, advised on national security policy, and run twice for federal office. Meanwhile, Rob is a field CTO of Global Cyber here at WWT, and before most enterprises were seriously discussing AI-driven threats, he was managing the intersection of regulated data, cloud infrastructure, and security outcomes in an industry where breach has direct human consequences. This is the AI Proven Ground podcast from Worldwide Technology. Let's jump in.

The Ground Just Shifted

SPEAKER_01 2:10

We had him on the show talking about mythos and the cyber implications. And, you know, one of the interesting things that he said at the time was to put the faith in the companies that were testing it out through Glasswing and other measures, and that the government should, you know, they should be looking to partner with these companies and not necessarily looking to override what those companies were testing out. And then lo and behold, as things do in AI, things change. And a month later, the government did override that. Access to Mythos was stripped away pretty quickly without much notice. Fable followed suit just a couple of weeks later. And by the internal accounts that we've had here at WT, it was one of, if not the most, AI security capable model on the market. I say that all just to ask you, Madison, give us a little bit of a lowdown on what's going on. Why is it turned into this such complex, dramatic situation? And what does that whiplash actually tell organizations out there about the state of AI?

SPEAKER_02 3:05

Sure. My God, we could probably have an hour-long conversation on just this piece alone, but let's talk about it from a couple of different perspectives of like what this means. And let's let's do it from like, let's do it from the policy government perspective at first. Okay. And let's also set the reality that we all are all learning as these models are being developed. And the models are turning into tools that we didn't even know that they would actually create. And realistically, that's how mythos like came about, right? We don't have to go into that whole monologue. But with that premise alone, then the US government has no idea really like how threatening these models are to national security. And given the fact that we are in a moment where we have geopolitical tension really, really high, then what does this mean for our critical infrastructure sectors, those being like water and utilities, utilities being electric utilities? And so what type of threat does that pose? Again, in this specific moment. And so the government's reaction was hey, let's stop and pause. Let's remove access to these types of models and say, hey, we need to understand them before we just give access to the broader public. Well we have to really remember like it's not necessarily about having the most sophisticated model. It's having a model that is connected to the right amount of tools or the right tools because it's able to pull together a number of different vulnerabilities and exploit those within a customer environment so quickly that, you know, it really doesn't require mythos. And those types of capabilities are already in our adversaries' hands. So I think there's a little bit of lack of understanding of what our adversaries are already capable of doing from a US government perspective and perhaps even like what we don't even know what they have entirely. From the customer side of the house, then what does that mean for them? I mean, one day they had access to what we believe to be the most robust vulnerability discovery tool to date to not. So, what does that mean moving forward for the way that you know they manage their own internal capabilities? So business continuity as usual. What is their new cyber business operating model looks like? And so it really kind of like changes the ownership of these models to, you know, is it really owned by the private sector or is it entirely going to be controlled by the US government moving forward? And, you know, we're all trying to figure that out. Even folks, you know, we're working on policy every single day of what we think the industry needs.

SPEAKER_01 5:49

Yeah, Robert, build on that a little bit, but also, you know, what does this all mean for a CIO or a CISO's office in terms of enterprise security architecture that that could be built on top of models that we don't control or from providers who could provoke access, or now, as we've learned, you know, a government or a state entity that can kind of pull that rug

The Fundamentals Still Win

SPEAKER_01 6:09

out from under us? Does that mean that current architectures were built on sand, or how should we position ourselves moving forward with this current um understanding?

SPEAKER_00 6:17

Yeah, you know, it's so funny. They're they're being called frontier models. And when you think about frontier exploration, you know, go back to your old history book days, and you had people with machetes and guns, and they were facing wildlife, and it was a little scary, but it was also exciting because they were literally finding and discovering new things every day. It's aptly named. Unfortunately, for the CIOs and CSOs that are trying to navigate that, it does look scary and it's it's coming in moment by moment, and things are shifting. But I do think it's important. And I one of the things I've I've had a lot of conversations just this week. In fact, this morning I was on with one. And we're just talking about taking a moment and breathe and remember that in spite of all the drama and activity that you're hearing drummed up, you're still in control of a lot of things that you can still execute upon. You know, we have been saying at WWT and frankly, the whole industry. No, Mattis and I had this conversation for the last three years. Uh, I don't think in a week goes by that the word cyber resiliency isn't used on a daily basis, if not hourly, in some conversations. So, in spite of all this shifting sands you see around you, some of the still anchors that you have been building upon in the past still apply. So, a lot of it is don't lose so much sight that you just hit the pause button on the activities that you were trying to move forward with anyway, on cyber recovery, understanding what those fundamentals are of knowing where your critical assets are, right? I mean, there's so many customers right now that we still go talk to, and they may have kicked off a lot of good efforts, but then they get mired down in the day-to-day. So it's really reinvigorating some of those fundamentals. Well, at the same time, you do have to gear up and think, look, as as we're navigating geopolitical issues and and national issues, as Madison was saying around this, you don't need to have fable. You don't have to have mythos. AI as a threat vector is out there today. And you do need to start to account for the, you know, understanding what the outside external attack surface looks like from an AI-based adversary, for example. You can start moving forward with that while all the rest of this is being sorted out. So it's definitely not a time to hit the pause and wait. It's a time to, you know, regather, prioritize your list on the fundamentals, and still go forward and think about how can I still embrace the future state as all this is being worked out.

SPEAKER_02 8:42

Don't want like our customers to think that they need to go build these models themselves because that's absolutely not the case. But I think what we have to realize that just because there's a new model doesn't mean that the one prior wasn't equally as robust for them to at least put in basic, I don't want to say basic, but to put controls in place. And so we're gonna have to get used to perhaps sure, not having access to the latest and greatest model, but feeling comfortable that we can still do the work to make our environments more resilient because those processes necessarily aren't gonna change if we think about the way that actors get in right now. The most sophisticated attacks truly still happen through compromise passwords, through subverting business processes, through open access ports, through you name it. It's it's the the adversaries are still winning in basic tactics. And so if that is the case, then we don't need the latest and greatest models. We just need to be able to scale our defense mechanisms quicker. Now, we can talk about, you know, if there is going to be better parameters around the way that these models are assessed with in partnership with the US government and some of our frontier labs. But, you know, that is not our responsibility here today. We need to just ensure that our customers understand what's going to happen if something is pulled

AI Needs A Harness

SPEAKER_02 10:18

from the marketplace.

SPEAKER_01 10:18

And Robert, I guess maybe a good way to restate that is the architecture of how you use AI is perhaps as important, if not more important, in terms of the actual AI you use. Is that relative fair?

SPEAKER_00 10:32

Not only is it fair, but it's scientifically proven. So it's it's not really opinion-based. I mean, you know, you can go out and take a look at the results that have been executed and what's what's called the harness, right? So when you think about I grew up on a farm, right? You throw a saddle on and how you'd have it on there. I promise you, if you don't have the harness on right, you only do that once. You fall off the horse, it hurts. So but when you think about that harness, it's basically what do you do, how do you leverage that power of AI? And we've already seen where the combination of not just agents, but skills, right? Really documenting and understanding what you want to accomplish. I kind of use the analogy of I could jump in and go really fast in a Tesla with FSD just by putting it in Mad Max mode and it'll get me there just with a destination. Or I could try to do that and race an F1 driver. That F1 driver is doing all kinds of amazing things with a whole different set of telemetry. He's gonna beat me. So, yes, if you have the right person at the helm of AI with the right harness around it, there's absolutely scientifically proven facts that from both exploits, vulnerability, et cetera, this has already been tested and proven. So I mean, it just kind of echoes Madison's point that there's probably layers of complexity behind where they're delaying some of this.

SPEAKER_02 11:47

There's a statement that I'm I'm really loving. And it is, you know, within the world where we're moving at machine speed, like discovery at this point is cheap. Like it's cheap, it's easy. What we are lacking to be able to move as quickly is the judgment, the execution, and the resilience. That's what's scarce right now for us. And so when I think about judgment, then I really think about like what, you know, how are how are we indexing vulnerabilities? Because I think that we've lived in a world where we've relied too much on like CVE or C VSS systems, like scoring, and we're not really thinking about that operational reality. And again, like I don't mean to be that doomsday like scenario person, but like I think about like how complex the environments are and how many dependencies there are, you know, within the world of critical infrastructure because of all the legacy systems, because of how complex they are with cloud environments, et cetera, et cetera. And so if we are thinking about, again, like discovery is cheap, and what we have to think about is that operational reality is the points that you were making. This, the fundamentals of cyber have it changed. That doesn't make it easier. It just means that we have to know our environment so much better now so that we can understand when a vulnerability comes in, we can prioritize it, we can understand, hey, do we need to just figure out how we can contain this? Because we're operating in a world where unpatched systems is basically the new norm. And so, what does business continuity look like when a set of systems are compromised? You know, how do we ensure that that blast radius is contained and isolated as quickly as possible? And even thinking about like deception techniques and really slowing an attacker down. And so it's just this whole new world that we're living in. And, you know, obviously the market has been moving in this direction for a while of operating from a place that you're already breached. But I think that now, given mythos, and I'd again I want to hear like your thoughts off this because of our conversations we've been having. But I mean, what was your conversation today? I feel like every CISO is thinking about mythos differently. Just yesterday I was chatting with one and he was like, we're just trying to patch so fast. And I I told him I was like, I respect that answer, but I don't think like patch faster is really the answer here.

SPEAKER_00 14:19

You you

The New Security Playbook

SPEAKER_00 14:20

you nailed it. I mean, the a lot of what we're talking the last couple of weeks has been this absorb, contain, recover is gonna be the new normal. And I love the way you put that. If you think about costs associated with everything, right? You do get to get a little bit in the spending wars when it comes to the complexity involved. Whichever model it is, it's gonna be burning tokens. And you know, the the AI Institute out of the UK recently published this fantastic finding showing a multitude of models. And those first initial discovery steps to get a landing zone inside an initial vulnerability are relatively inexpensive. But when you start to move up the stack, you start to hit into you know the millions and tens of millions of tokens. And I think that is our opportunity window, right? Like from an absorbed contain, we've talked for years about micro-segmentation, but now let's take it to the next level. What let's segment once they're in there? You mentioned deception. What if they come in and now we can occupy them with some deceptive technology, learn what they're trying to go do, contain them, take that learning, apply it through the rest of our environment so we can see as they're trying to go move laterally other places. But also, how can I now leverage AI from uh AI in generic terms, but also automation? For example, what if I rotate API keys now or secrets? If I know that's how they got in, now that I got them in here, I know how they came in. I'm learning more about my enemy. It's it's Sensu, it's the art of war, right? It's some of these fundamental principles of defense are just playing out differently. And that's what I was talking with today with a major retailer. They're like, look, we're trying to figure out, frankly, they feel a little overwhelmed because they're they're still trying to figure out how you do MFA when some of your employees don't carry a corporate phone. You don't want to pay for them to have a device that's MFA, right? So conversations are still going all the way back to some of those fundamentals to, but on this other side, how am I going to recover once I see them in here? They want to know how to change their logging strategy to now see further down the stack into some of those messages that could now be indicative of a of an AI-based adversary coming in.

Your Competitors Aren't The Problem

SPEAKER_01 16:29

There's another tension kind of brewing over on the other side, which is, you know, you can restrict mythos, you can restrict fable, and that can buy us some level of time from an enterprise standpoint. But there are still open weight, open source models that bad actors can get them their hands on. Uh, how do we account for that friction? Is there a little bit of a ticking clock that we do have to move with extreme urgency? Or, Madison, is it more close to what Robert is saying? Like, hey, take a deep breath, focus on your fundamentals, and that's going to enable you to move as fast as you practically can go.

SPEAKER_02 17:01

Our adversaries already have incredibly sophisticated models. So, respectfully, I don't think that it really buys us any time. And again, I'm not the doomsday scenario person. I'm not here to do scare tactics. I think it's just understanding the operational realities in which we're living in. Now, absolutely, I'm with Robert. Like, let's all take a breath and figure out like what does this mean in the new normal? But realistically, again, it's still going to be about identity and access management. It's still going to be ensuring that we have visibility into our environment, understanding what's normal, what's not normal, being able to, you know, do those types of investigations. It's just now we need to do it at machine speed. And so the there are gaps in what right now is technologically feasible. You know, yes, there are AI tools and existing products that have been on the market for a long time that are starting to leverage AI. But I say this with all respect to, you know, my homies and my defenders within the cyber world, but marketing and capability doesn't always match. We need self-healing environments. We need to understand, we need to have tools that understand those downstream impacts. We need better asset inventory and asset discovery capabilities. And, you know, I don't want to get too long-winded because I definitely want Rob to like Robert to chime in here, but you know, I do think there's like this shared responsibility model that we should chat about at some point because like our customers can't continue just to absorb the brunt of vulnerabilities that they don't necessarily have control over. Like they're just inherent within their environment.

SPEAKER_00 18:47

We let off saying patching used to be the primary response. That's no longer the primary response. It's it's huge. We got to get better at it, but you have to do other buying time mitigation scenarios that you can go leverage. And through that, you're not gonna be cutting across organizational silos, frankly, into teams that in the past maybe never had an actual, I'll call it active incident response role in some of these meetings, right? So you're you're really gonna be really modifying the business. Now I say that because AI brings a lot of capabilities together as well, but your your model is always gonna be sitting within your holistic infrastructure. So we want to make sure that we're addressing it holistically. And I I think that's where we're at this cusp of having to go move rapidly. But a huge part of that is still execution on those fundamentals cannot be dropped. This just has to sit within those fundamentals as well. So it's definitely causing a groundswell of workload and workforce transformation that I think we had a lot more time to do that when we made the move to cloud, right? Like people had cloud center of excellence of a three-year plan. And now we're like, well, this is gonna be an AI center of impact, and it's happening in the next three months. So you better be getting people together.

SPEAKER_02 20:06

Yeah. I mean, I recommend at this point that every single organization, yes, obviously has that AI center of excellence, but every single day there's new use cases developed and really thought about. And there's almost like weekly stand-ups happening of like, okay, what new use cases can we develop? Because, you know, again, if we go back to that statement that discovery is cheap, but execution and you know, the orchestration pieces are what's missing, then to me, that means that we need to be able to leverage the ability to create all of these like agentic agents to be able to do that work for us.

SPEAKER_01 20:42

So both of you now have pushed back on the idea of this can be solved by or we should be just patching faster, right?

Stop Trying To Patch Everything

SPEAKER_01 20:50

Both of you said that's not necessarily what you see as a smart way forward. Why, Madison, is that not a solid operating model?

SPEAKER_02 20:57

So the idea of the Mythos model was they were the idea was to create the most secure coding processes in the world. Let's just put it in that way. Well, to to be able to, in order to do that, then you have to go back and validate and find vulnerabilities in that code. So then a tool was then basically created, a model that can find vulnerabilities, you know, I think at 70%, 60% quicker than anything that we've ever been able to discover. And so what you're saying, what we're saying is the number of vulnerabilities in a single system that you know have not been enabled with AI are probably riddled with vulnerabilities. And it could be 200, 300. And it could take a team to work through like two or three vulnerabilities, especially if they're like critical vulnerabilities, because there is that lack of understanding. So let's let's talk to that process. Let's say I don't want to pick on Microsoft, but we'll just say Acne Company. Acne pushes out a vulnerability and says, Hey, this is a vulnerability that we found. We're marking it high just because of the operational parameters around it. Well, that the customer basically has to say, okay, let me validate this. Let me see where it is in my environment. Does the patch actually work in my environment? Is it going to brick anything? So we don't have time to validate and test those things, right? For every single new vulnerability. It just doesn't work. And so if we can't patch fast enough, then that's where and why Robert and I are saying, okay, if we're going to have holes in our environment, how do we contain so that again, for a major electricity company, we're not seeing massive blackouts in a major city like Atlanta or a hospital been turned off. We want to see maybe like a house, one house, you know, lose electricity. Obviously, we don't want that to happen whatsoever. But it's again containing that downstream impact. And so ensuring that that adversary, either that we're able to find them at the quickest amount of time, or again, that contain piece.

SPEAKER_00 23:06

Yeah, uh Madison, I think this is where we're seeing maybe one of the most encouraging explosions of innovation as well on what these mitigating technologies are.

Buy Yourself Time

SPEAKER_00 23:16

You know, I'll I'll pick on the one that we've already been using for several years. You think from the web application, right? Maybe you'd have an external web pen test and you'd find criticals and be like, well, I have to patch 9,000 endpoints, or I put one rule on a web application firewall, and it buys me time. So on my normal quarterly or whatever, you didn't have to go immediately touch them with urgency because you had a mediating control. So you're buying time technologies. And we're starting to see more and more of this. And I think you're gonna start to see this. We've actually been experimenting again in the ATC with some of these with other infrastructure providers and ways that they can go through and mimic that WAF-like functionality inside of the protocol layer. I suspect we're gonna continue to see more and more involvement there. And and the critical piece of this is it's no longer driven just by that vulnerability scoring, as you were mentioning. It's like, no, these are actual exploits that I know that they can be accomplished. And how can I go leverage this right now to immediately affect that change? So I I every place that we're going from this is gonna help that again, that term cyber resiliency. This is all gonna help us figure out how to be more cyber resilient. So from an from an optimist, and you don't hear that often from the cybersecurity world with this, but I do think we're gonna, you know, we're starting that camel curve of having so much change coming in. But if we can look on the other side of this out, I do have hope that with some of the tooling that's coming to the table and some of the I'll call it the new normal, that I think we can become more resilient altogether.

SPEAKER_01 24:46

Robert, I am I am interested. You're talking about buy buying time technologies. That's not necessarily a term that I hear a lot about. Is that is that relatively new? Has that been around for a while, or is that specific to kind of this AI scenario we find ourselves in?

SPEAKER_00 25:01

Well, that's a great question. It this is a it's like you kind of forget about some of these tools you have in the toolbox. It's something that's been around a lot of folks, frankly, before we went all in on some of the massive telemetry and the sim collecting of logging, a lot of organizations would create honey pots and things like that just to help identify, and frankly, a lot of times it wasn't necessarily to identify a bad actor as much as it was to identify a performance-related issue or a fault-related issue, right? Have some, you know, we we use that term canaries of the coal mine for a reason. It started with canaries and coal mines. So this concept has been around for a long time. You know, how can we find some of this? But it's, you know, what is old has been made new again. So we're starting to see not only canary and the coal mine, but imagine setting up some domains that could be accessed to learn what that attacker is doing, actually set up some dummy domain controllers in that space, even some dummy databases, right? So I think you're probably going to start to see, especially with our critical infrastructure sectors and global fi, I think you're gonna see a whole lot more folks trying to leverage this so that they can learn what's happening. And I mentioned before that recover, right? Like having nowhere they're at. Now I can segment and I can start to figure out what are some of my automated recoveries if I have to go through and do a whole blast and rebuild, right? Like what does that look like in that space?

SPEAKER_02 26:25

I don't think that it's a new chain of thought that we've been kind of like walking through. It's just again, kind of the speed. There has always been kind of this thought process that there is defensive latency with any type of new like patch that rolls out or a new vulnerability. But what's changing is, and if we just again kind of think about, you know, so it's kind of three layers the way I think about it. It's this defensive latency, and then it's a new vulnerability. And so it's that awareness, that context, meaning, you know, what is the asset? Who is the owner? What's that reachability? Can we confirm that? And then there's this containment window, and then there's the remediation. And between awareness and context, that's your risk window. Well, it's really your awareness all the way to remediation. That right there is your risk window. And so what we're trying to do is understand what our defensive latency is in this new world with AI, and how can we reduce that risk window as much as possible? If it's not going to be patching, then to your point, Robert, you know, what can we do to really enable our customers to learn from some of these new vulnerabilities being exposed, but also some of the new tech tactics where we'll see more honeypots being used, deceptive techniques being used.

SPEAKER_01 27:48

So the playbook exists. It's about closing that risk window before adversaries can move up the stack.

The Accountability Gap

SPEAKER_01 27:54

But there's a harder question underneath all of this: who's actually accountable when a breach happens? And is the current answer still fair? This episode is supported by Cloudflare. Cloudflare's global edge network protector applications, users, and networks with unified statistics security, DDOS protection, and AI-powered threat defense, blocking billions of cyber threats daily. Secure and accelerate your digital business with Cloudflare's cloud native security platform. I mean, organizations are still being held, you know, broadly speaking, accountable for breaches, even if they occurred upstream, whether through a product, a platform, or you know, an open source model or whatever it might be. Is that allocation of responsibility still justifiable in the age of mythos, AI, however you want to term it?

SPEAKER_00 28:46

Well, that that answered very greatly based on who's in the room if you ask them that, because many people, you know, Madison's used that term twice now of the shared responsibility model. And, you know, I I I love that word accountability because it really needs to be the who's accountable model in this new world order. And when I say that, we have traditional silos, for example, application dev teams that are used to moving really, really fast. And some of the things that they're going to go leverage, they may not understand what that's really exposing them to. I've not brought it up. We could have an entire podcast just talking about what third-party supply chain is in the AI-driven world, where now every SaaS-based provider has their own AI-related chatbot. And, you know, people are so excited because they can get results so much faster and they don't have to poke through menus. But similar to when, you know, for the longest time, people would just scroll through a user agreement and click accept, not realizing that their data was the product. This is kind of what we're seeing within enterprises now with some of this third-party supply chain. So bottom line is ultimately the the the big C's, the CIO and the CISO, when it comes to who's going to be accountable, the business is always going to go back to the big C's and say, why didn't you catch this? Why didn't you catch this? And I think, you know, Madison, echo your point. That's why we can't go to quarterly conversations on how AI is proceeding through the organization. We are seeing this literally being stood up in some of the industries I've been talking to, more similar to a war room where it's a daily update, a daily stand-up, where do things sit today? And we're seeing things move at a pace here that frankly is unrivaled. In fact, I just saw the World Economic Forum 2026 just published. Cyber risk is now the number one, not just for business, but the number one global risk that we are facing, bigger than everything else. And if you if you pause a second and stew on that, there's a reason why you should be talking about this and listening to these podcasts, and everybody has to be educated on it. You can't say that's just for the IT people or the cyber people. It literally is going to affect us all. So I think that's why you're starting to see such a huge focus on this across the board.

SPEAKER_01 31:05

So accountability is real and it's moving up the org chart fast. If it's the number one global risk, then the question isn't whether to act, it's what acting well actually looks like. Madison paints that picture for us. Madison, what what does a sh what does good look like here? What does a what does a meaningful shared accountability scenario look like, whether it's between organizations and their their vendors or even public-private partnerships? What does good look like in terms of shared accountability?

Security Can't Do This Alone

SPEAKER_02 31:35

Like the enterprise or a customer is always going to be held accountable to a breach. There's just no way getting around it. But when I think about a shared accountability model, then like who are the players? So, in the way I think about it, it's our customers and our operators. It is the OEMs that are building big types of equipment that is operating in a manufacturing floor, that is operating within a hospital that are saying breaker open, breaker closed. It's the cyber product companies, it's our large cloud providers, like our Microsoft's, our Googles, AWS, you name it. And the other side of that is, you know, basically the US government as or name that standards organization, like a NIST or a miner, a MITRE. All of them have a responsibility within their ecosystem to help build overall resilience for our customers at the end of the day. And what does that look like? For me right now, where I think the biggest gap is, and again, I'm not here to throw stones. You know, what we love about living in the age is like free innovation and free enterprise. And I certainly don't want to slow it down, but there are regulatory standards like NIST that our financial sectors have to abide by, that again, our critical infrastructure sectors have to abide by, but there haven't been defined standards that are truly hold our OENs accountable or our large like manufacturer or product manufacturers. And again, if there's only three or four of them or four or five of them that make this big level of equipment within the space, then why aren't we saying that, hey, TLS is basic? And if you don't have just, you know, things not operating in clear text, like that's negligence in my mind. If you don't have secure processes, secure coding processes, that is negligent. Right now, what's been holding them accountable is really the market. But I think that we're getting to a place where we're so used to companies getting breached that we're almost like, okay, well, I've just got to absorb the risk.

SPEAKER_00 33:45

And I don't think that it's fair and yeah, not to overcomplicate it, but uh, I just read an article this morning that was talking too about open source systems. And, you know, the entire industry has benefited from these over the last 15 years, and yet they also are the ones where they're finding these massive backlogs in vulnerabilities. And now you're dealing with, frankly, what's been a a best effort community that is overwhelmed, like they can't even begin to keep up with what they're seeing. So it's it's like these bookends where we've got massive infra providers that you know. I think about Glasswing and somebody's like, yeah, there's gonna be a new tolerance, I think, that the industry is gonna say it that's not acceptable now. I've I you've got to prove this. There's no reason. If anything, it's brought so much attention to bringing software quality up from just a feature to also a security perspective. But then that that accountability quickly wanes when you turn to the other side of it, where so many people are going to GitHub and using an open source. And you're like, well, if that's not secure, who do I blame? But I think that is gonna 100% require a policy-related thing to frankly almost pull funding from everybody else to go say, we we all use this. Everybody uses this, we have to improve that today. So I I think I think that's we're gonna have to start to see some dramatic shifts very soon.

SPEAKER_02 35:13

Yeah, and I think we're seeing the market go in that direction, you know, to to give Google like a shout out here, and you know, I don't get paid by Google to say this, but you know, they they bought whiz, right, to help them be able to say, hey, maybe we are the most secure cloud provider in the world. You know, maybe, but it at least I think is is moving in the right direction.

SPEAKER_01 35:33

Given everything we've talked about, there's a lot happening. There's a lot on the plates of security teams, and the business always wants to move faster and faster and faster with AI.

The Defender's Advantage

SPEAKER_01 35:44

Madison, maybe assess for us where are we at in this kind of chess match right now? Are we starting to see light at the end of the tunnel? Are we just now in the early phases and there's a lot of game left to be played? Where are we at and what do organizations need to do to best position themselves to win the game, no matter how much time is left?

SPEAKER_02 36:03

I mean, we are if there, if this is going to be a walk, crawl, run type of analogy, I think that we are like crawling right now. And I don't mean the pace, I mean in the level of like just understanding of the potential that AI is going to bring, both from a defense perspective and like an exploitation perspective. And so I think that we are just seeing like the very tip of the iceberg of what the heck is freaking possible in both of those realms. And it both scares me and excites me. But right now, I do think that defenders have the upper hand, or excuse me, opposite. I do think our adversaries have the upper hand right now, where AI is not to the maturity level yet to really empower our defenders at the scale and pace in which that they need.

SPEAKER_01 37:00

Robert, if if if the bad actors have the upper hand at the moment or in this inning, what are you know, what are some of the two or three priorities that we have to do to start to balance those scales?

SPEAKER_00 37:10

So uh unfortunately, a lot of this is revisiting things that we've been talking about with cyber resiliency over the last five years. This has just put it into overdrive. And I think, frankly, it kind of comes back to there's not because AI is a very complex adversary, there is not a silver bullet that's going to help you across the board. It's really sitting down tactically and figuring out multiple parallel threads that you can go adapt to to continue to deliver on those cyber resiliency exercises while you're also trying to figure out how to embrace AI from a from a offense, uh defensive capability, right? From an AI response. And I even tell people as you're trying to do this patching, let's go do better automation of your patching. Let's at least get better at that. And in the process, don't forget to build an undo button because you will hit, you'll stub your toe on that patch, and you better have the undo to go restore service. Because if you take it down, whether the adversary does or you, the business is down. They're not going to want to hear that. So, you know, just thinking about some of those fundamentals. So, yeah, there's no, it's not a simple answer, and yet the process itself is not overly complex. So you sit down and do the evaluation.

SPEAKER_01 38:23

What does that mean about model dependency in terms of is is model selection and analysis, is that now as much of a you know a business continuity question as it is kind of a technical question, do you think?

SPEAKER_00 38:37

Frankly, I think some of it could just be them trying to absorb the pipeline of change, right? Like they probably haven't had time to go through and figure out how exposed they are to some of these new models. So I think there's definitely a lot of complexities and hate to say, even some politics perhaps around some of those aspects.

The Question To Take Back To Work

SPEAKER_01 38:55

Okay, thanks to Madison and Robert for joining today. What Mythos and Fable exposed is bigger than a disagreement between an AI company and the government. They exposed a new category of enterprise dependency. So the question that you need to take back to your organization is if a model disappeared tomorrow, or if an AI-enabled attacker arrived tonight, could we still absorb the impact, contain the damage, and keep the business operating? This episode of the AI Proving Ground Podcast was co-produced by Nas Baker and Kara Kuhn. Our audio and video engineers, John Novlock. My name is Brian Phelps. Thanks for listening. We'll see you next time.