Video: AI & the Threat Landscape: How the C-Suite Are Governing AI Before It Governs Them | Duration: 2772s | Summary: AI & the Threat Landscape: How the C-Suite Are Governing AI Before It Governs Them | Chapters: Welcome and Introduction (18.205s), AI Security Challenges (69.835s), Field CISO Insights (133.4s), Shadow AI Statistics (256.66s), AI-Driven Vulnerabilities (402.435s), AI Defense Transformation (520.995s), Shadow AI Risks (683.33s), AI Security Boundaries (841.15s), AI Governance Visibility (1022.035s), AI Policy Enforcement (1262.07s), Measuring AI Usage (1447.575s), Shared AI Governance (1669.965s), AI Governance Advice (1772.625s), AI Risk Frameworks (1950.145s), Security Primitives (2116.78s), Closing Insights (2317.7s), Session Wrap-Up (2566.56s), Rippling AI Overview (2612.78s), Closing & Raffle (2717.905s)
Transcript for "AI & the Threat Landscape: How the C-Suite Are Governing AI Before It Governs Them": You seem, yeah, done deaf scheme day. Just doing what I do. Alright. Welcome, everyone. I'm Matt Lombardo, and I work on the IT side of things here at Rippling as our strategy and community lead. As you all know, October is cybersecurity awareness month. And this year's, CSIS theme is securing the next two fifty, a call for every organizations, to strengthen defenses. The pressure on that mandate right now comes from two directions at once. Attackers are using AI to move faster, and employees are adopting AI tools faster than security teams can track them. And that's what we're here to dig into today. You know, with these three security leaders, you know, who deal with both sides of that problem every day. Adrian, you joined Rippling as the as the chief security officer after Android, Atlassian, and Tools for Humanity. You know, just give quote, folks a quick version of your role here at Rippling, and what's the biggest security issue on your mind today? I suspect we're not that unique. The biggest issue on probably everybody's mind in security is how AI is gonna affect their organization. Some of it is how do we incorporate new technologies into existing programs and make sure that those programs work as well as they possibly can. In fact, can they work orders of magnitudes better than they could six months or nine months or even a year ago? And then the second is there's this dark, ominous cloud of the future, where we wonder what's gonna happen as those capabilities become more and more sophisticated. We've seen a massive shift in the last year. There's nobody who's saying that that's not gonna continue, and from a security standpoint, that's, that's top of mind. Awesome. Thank you for sharing. And, John, you're in a different seat as a field CISO, you know, which means you're in front of security leaders across industries every week. You know, tell us about your role and and what it's given you visibility into. Yeah. No. Thank you. So, yeah, it's it's a little bit different than when I was on the enterprise side. I was a two times CISO in health care financial services. And so the conversations, it's it's really been a unique opportunity for me to think... To talk about some of the strategies that I deployed when I was in the seat versus now being more of an adviser and helping, organizations develop a strategy for it. And it's there's there's a lot of trying to figure out, what are the detective capabilities, what are the preventative... Or what what are the protective capabilities, but then how do you now factor in preventative type capabilities? So, it's it's been a very unique conversation, but I think the perspective, from how we use AI here at Chainguard offers a lot of insight into that coupled with that form of experience on the enterprise side. Awesome. Thank you for sharing. And, Tahala, you recently joined Vercel as the CTO of security after seven years as a CISO and CIO at HashiCorp and running security for all of IBM. You know, what pulled you toward a role that's specifically about securing, the AI era? Yeah. I think I've always been in open source dev tools infrastructure companies, and Marcel sits at the nexus of, capturing the most modern developer. We've contributed a ton on open source on products like Next. Js. What what drew me here was just the, the build out and interesting challenges, from practitioner angle on how developers actually, are using, modern dev tools to build AI products and agents, and very privileged to be here to help them think through securing their journey. Thank you. So before we get into it, you know, just a few numbers to kinda set the stage about what we're talking about today. Right? We... With Shadow AI, which are AI tools that our employees are using without securities knowledge, now touches an estimated 66% of office professionals, according to PagerDuty's 2026 Shadow AI workplace survey. And IBM's cost of a data breach report found that Shadow AI, linked incidences jumped from twenty percent to a staggering forty three percent of AI related breaches in a single year, adding to an average about $670,000 to the cost of a breach. And an Okta survey found about sixty eight percent of CSOs have been seen at least some unauthorized AI use of inside their own organization. But Gartner puts a share with a formal, you know, detection program at just 34%. And at that at that same pager duty survey found that 43% of employees have shared work emails or other sensitive correspondence with public AI tools that aren't connected to a company systems. And, you know, that's the gap that we're here to talk about. You know, how are how are attackers are using AI against you, you know, how your own people are creating exposure without meaning to, and what it actually looks like to get ahead of both of those things. So, you know, John, I just wanted to start with you and just kinda, you know, what changed in the last twelve months in how attackers are actually using AI against the organizations that you work with? I think the obvious answer here is is the use of Frontier AI models because it's accelerated, the attacker's ability to, discover vulnerabilities, but also improve the quality and increase the scale of their attacks. That has been the number one thing and that... You know, although that's, you know, just been in... Once that this has occurred, we are just seeing that how do we... Organizations are trying to figure out how to get ahead of that. And, you know, detective and protective mechanisms are are great in in place, but, you know, the existing security tooling and technologies don't exactly meet, the need when it comes to, you know, the use of AI by these attackers. Agreed. And and, Tahoe, has AI shortened the gap between a vulnerability, you know, being disclosed, and it being exploited. Like, has AI shortened that gap? Or, you know, what are what are you seeing in the wild? Yeah. I think the two data points I'll share, one, the traffic on the Internet from agents has now surpassed traffic by humans. So that's just a indicator when agents are running on behalf of users and running wild. The the the... They don't work, in human hours. They work around the clock. So that's one. And, John, I already do the user Frontier models. I don't think it's, like, just the Frontier models. It's the... The the way I describe the current state is it's the best time in the industry for offensive security research. It's the worst time in the industry to be on the defense side. And the reason I say that is the availability of models, the the the velocity of vulnerability research that has one being commoditized and two in the hands of a lot more people. And, yes, you still need skills, but the the the the models are doing a lot more hard work for you. So I think both in terms of the the velocity as well as the scale of vulnerability discovery, it's impacting just the fundamental assumptions about how open source software used to work, the kind of, protections, people believed, you know, come from software, and those assumptions are just breaking, significantly. So what we see, I think, when these agents are also used to exploit software is just the the velocity and persistence of, how these attacks are conducted. Wow. And Adrian, you know, where is AI making your job as a defender easier, and where is it purely making the attacker faster in your opinion? I'm fascinated by what people mean when they say that something is hard. I think what I'm seeing is traditionally an awful lot of time is spent by a security organization knowing that a risk exists or knowing that a vulnerability exists, but then going through the process of triaging and prioritizing and stack ranking and negotiating and debating and then maybe fixing. And so in that sense, while it is strictly speaking more difficult to defend now because things can be exploited more quickly and can be discovered more quickly, all of those intermediate steps are basically no longer relevant. So in some sense, it's actually easier to defend because I can spend time going from vulnerability discovery to resolution immediately, skipping all of those steps that had been put in place previously because there was a lack of alignment on how important it was. So I think, in that sense the speed with which we're seeing AI demonstrate that it can go from proof of concept to reality in minutes, means that the engineering process has to do that as well. It means that companies have to react that quickly as well. Now if you're in an organization that can't react that quickly, then you have a problem. I'm fortunate that I am in an organization that can react that quickly and we are reacting that quickly. And so, in that sense, it's just an incredible opportunity for us. It's I won't say it's easier to defend, but, the shift has gone from sort of organizational discussion to just, like, let's fix it and move on. And I think that's what's gonna have to happen across the entire industry. I don't know if it's this month or next month or over the next year, but it's hard for me to imagine that it's gonna be more than a year, given how quickly these things are happening. So and then and then I think on the backside of that, my expectation is we'll have relatively robust security models that are actually implemented correctly. Whereas right now, we have neither particularly strong models nor particularly good implementations. So it's a little bit of a crucible, to to prove that these systems are actually designed well and implemented well. Wow. So now I kinda wanna shift from, like, what we're talking about with ex... External attack... Ex... Attackers to more of, like, internal. Right? Like, the harder to harder to see risk where employees are adopting AI faster than kinda maybe we're tracking it. So, you know, Tala, what does shadow AI look like inside a real organization today, and what are people actually doing that security maybe doesn't know about? Yeah. I'll start with I'm very happy and privileged to be in a company that is building the agentic stack so we can enforce our own process and tools on our people to use environments correctly. But just speaking to my peers and seeing what's happening in the industry, I think the first big trend with the the with the advancement of AI was software became free. When I say became free, it's like it... The the the availability of coding agents and how many more people can now build applications, they just increased vastly in every company. You are now able to... A lot more non developers are able to generate apps and make business apps, and that's the transformation you're seeing in many companies. And in in some ways, the the problem of shadow IT isn't that different from the traditional shadow IT problems. It's unsanctioned software. It's which apps are, what's the SDLC, and how do you build the right guardrails and controls, and which environments to be used, and what's an approved and non AI approved use. So I think the fundamentals are largely the same fundamentals that if you had them right, you would govern IT... AI better. If your fundamentals weren't that great, then I see people struggling. I do think there's some uniqueness in the, in the security issues that AI agents are creating. You know, a lot more people when they're generating software, the agents are running locally on their machine and is connecting to business apps. So, I think companies do have to think through what are the right guardrails and deployment platforms and and, integration patterns and launch patterns, on internal versus external agents and, criticality of their business apps that, some companies are much more mature than the others. Like, on our side, we enforce those patterns and launch criteria and human in the loop approval systems. Some companies are just early in their AI adoption and still learning through. That totally makes sense. And, Adrian, you know, from from your side, what's the most surprising thing you found once you got real visibility into how people are actually using AI day to day? And, you know, how has that improved Rippling's AI governance? So I think the the most interesting part to me is how AI doesn't really seem to know where any of the boundaries are, and it moves pretty fluidly through them. So I'll give an example that we've encountered pretty regularly. Humans know when they're using an application. Right? You open your browser, you click around, you log in, you interact with it. Humans know that that is distinct from accessing something that's sort of local on your device. And there's some confusion. There's people who think that everything is part of Google because that's the first pair that what's on there, you know, they don't realize they're using browser in the browser. But for the most part, like, we differentiate those are two different things. We're finding that that segmentation isn't intuitive to the AI agents and so they blend things together in a way that leads to curious security outcomes. The probably the most obvious example that we bump into is agents love to pick up tokens and just use them. And they don't know that they shouldn't be using tokens that were in the browser somewhere else. They don't realize that the security implications of that. And so I think there's a a fairly large need to rework how identity works in most enterprises to get to a point where, agentic workflows or AI based workflows or AI tooling that's being built, maps to the sort of security expectations that preexist those workflows. So that's been the most surprising thing for me, is how many times we see an app and then we go talk to it, sometimes it's a developer or sometimes it's not a developer and we say, do you realize that this thing you've built grabs all of your tokens and, like, sends them somewhere else? They're like, no, I would never do that. That's terrible. Like, well, it did, and you're gonna have to rotate all of those now. And that conversation is happening, like, every day. Wow. And so a lot of the security boundaries that didn't get violated because people just didn't think to violate them, agents don't even know they're there. So that's that's probably the big and and that's part of why you need to have the visibility into, like, who's running them, where are they running, and and where are they connecting. And so the sooner you put that in place, like, I'm sure Vercel's got most of that locked down because they're building the tooling. Or is it, you know, we we are building our own for the rippling part of the stack, but the daily, sort of endpoint part of the stack, we're using a a wide variety of different frameworks, and and and as a consequence, there's there's complexity and and we discover new things every day. Your token story just made my heart jump about three three beats faster. Yeah. It's fine. But, John, I mean, John, and, you know, is the goal realistically full visibility, or is it, you know, kinda... Is it, like, high risk, or is it, like, risk based triage? And and how do you decide where to draw the line? Well, I think the goal is, more about effective AI governance. And governance is about visibility. And, you know, most organizations don't realize that they have some level of natural visibility through either their Caspian solutions or their secure web gateways. Some of those network security capabilities, they have that initial level level of visibility, but then it becomes more about the conversation around, okay. How much do we need in place in terms of guardrails? You you can't go full bore, lock everything down because now you are disrupting the business. And that's not something that, you know, that's the balance you've gotta strike. Right? Is, how much is enough to to be able to protect the organization. But in order to do so, you've gotta have that visibility. You've gotta know and understand what type of guardrail, if anything, you need. You know, as Adrian just talked about, you know, I... To... The tokenomics, you know you know, behind all of this. You know, how much is this stuff using? But I think there's there's another component to it of what visibility gives you is fourth party risk that most people don't think about. Because that third party solution that they're using, they need visibility into what is that, application, what what is its purpose? What... How is it being used? Because the intended use or what it was designed for, is not necessarily how it actually gets used because there is, in in many cases, a fourth party risk because some of those AI solutions are built or even just applications that use AI in their application maybe using... You know, so now you've got that third party application that's using this fourth party application. And, you know, I think of, situation I had in in my prior role whereby you had people using some of these, AI tools to write emails and to write memos and and generate reports and to analyze reports and do all these things. What they don't realize is that there's a subprocessor. There's there's another AI tool underneath that, and one in particular I found through visibility was that, one of those tools they were using to to generate some reports and do some analysis was actually using, DeepSeek as another level down below it. So now what... Now you've got two steps of terms of conditions and and, usage policy that you've just signed into that you don't have visibility into. So, that's why I say it's about AI governance, which is visibility, give... Giving you insight into what's being used, where those... What other applications are being connected into those applications. You know? So, for example, you know, like like with Rippling, you know, a user may be using it to kind of analyze, let's let's just say, salary data. And, you know, but they aren't realizing that they've now connected another app into Ripley. So now they've expanded their attack surface. So that's where security has the opportunity to make them aware of that and and talk through, okay. How are we going to address this? Because we're expanding the risk to the organization, and that's not what we want. We're not here to try to disrupt your your use of this because, you know, shadow AI is is not... It's never been about malicious, intent. Right? This is just an everyday employee trying to be more efficient, more effective at getting their job done, but they don't realize how they're expanding the risk associated with that. So in order to, understand that risk, you've gotta have visibility, and that's the governance part of this. Oh, thank you, John. That that... You're going exactly where I kinda wanna go is, like, how do we look at this whole thing holistically without kinda killing the velocity? You know? So so, Adrian, with that, you know, how do you write an AI usage policy that people actually follow, and how does that help improve organizational AI governance moving forward? I I think the, I don't know if we know yet. I think, I think we're gonna have to experiment here for a bit to, to figure out what that looks like. I will say most of policy enforcement in most enterprises at this point basically comes down to do employees do the right thing and does it turn out that their understanding of the right thing maps to roughly what the business thinks they should be doing, right? Because people don't read the policies. People don't check whether the action they're doing today is consistent with those policies. And the extent to which an organization is able to have technical controls mapped to policies is very, very small. And so we're just kinda YOLO, and lucky that for the most part, humans do what we expect them to do, which is not how agents work. So in that sense, I think the great reckoning here is how quickly can we get policies that the agents themselves are able to consume and are able to be constrained by, and how do we put those guardrails in place so that the agents end up being forced to follow policies in a way that humans never had to be forced to because humans just did it sort of naturally. You know, the the example that that I bump into is like, I know that I shouldn't do the same action 10,000 times in order to get a comprehensive list of behaviors across all employees. An agent doesn't know that. And so it's more than happy to just go off and do something 10,000 times, aggregate it, in a way that a person never wouldn't. So I I think that's the hard part, is mapping, sort of intuitive expectations that humans use to constrain their behavior and making sure that they can be enforced on, on agents. I don't think we're gonna be successful at stopping people from adopting automation everywhere. It's just too easy. It's too flexible. And so then it becomes for every use case, how can we point them to a set of tools that they can use that will work well, and that will follow the the policies and and and will implement, things that are consistent with the controls that we want inside the enterprise. So for us, it's been how can we move faster than our employees at making AI accessible to them, and telling them, hey, you should be using x. So they don't have to, you know, go online and say, this is a tool that I should be using, because we've already told them what tool they should be using. That's a great point. And, John, what does tracking AI usage, actually mean operationally, and and what are you measuring and how? Well, I think what it what it means operationally is what are people using? And going back to a point that Adrian just made, understanding what they're using and redirecting them, providing some level of coaching and guidance in terms of, what they should be using, but also giving them some understanding of why. And so being able to measure, the increase in the number of applications that are used because, you know, we we we talk about approved, unapproved, sanctioned, unsanctioned, whatever the term is used in an organization. You you can't keep up with that because there is... I don't know how how many, AI, pieces of software are out there today, but I I remember seeing that number grow, you know, on an average of 10 to 15% almost daily because there's there's some new, fancy feature. And I think, you know, open source software certainly made that, a lot easier for for people to create new versions of things and put a different spin on it. And so being able to, you know, measuring just how much you've increased the number of AI apps that are used in environment, that's that's not a useful number because it's it's going to make you feel like you've got a bigger problem than maybe you necessarily have, but it is, again, understanding what's being used, understanding what level of... What type of information is being put into these. You know? What are what are people using in those prompts, as it goes through? Understanding what... How many connectors you have between those different applications. I think that's the first thing to measure because that's that's where your risk is. That's where the expansion of your attack surface lives. So now that's where, from a security standpoint, you're trying to figure out how do I not disrupt that that innovation? Because everybody... You know, AI is being used in the name, but, you know, innovation is being used, in in, concert with AI. You know, they're they're kinda synonymous in organizations today. It is everybody goes, well, we've gotta be more innovative and, you know, we've gotta be creative. We've gotta create competitive advantage. We've gotta gain operational efficiency, all these things. So it is... I'm I'm trying to measure what is our ability to enable the response to use and secure adoption of AI. So things like how many diff... How much how much date... The volume of data being uploaded? How how many visibility into... Or measuring those number of... How much are we spending in tokens and being able to gain an understanding of that? Because some of the tools we have now in our in our cloud environments, we... We've tried to use to do that. We... We've tried to repurpose that, and I don't think that's that's very effective. It really kind of gets more confusing. You create more of a defensive, position from the business than what you want. You really wanna be more collaborative, in nature and in how you measure that. So I think it is being able to understand and measure, in terms of our business processes, what... Which business processes are utilizing AI the most, and and what's the volume of that transactions that are going through, but also how much are they connecting to other applications? Because then you can start to figure out from a guardrails or governance standpoint, what is the risk and where where do we need to prioritize our focus? That makes a ton of sense. And rounding us out, you know, Tala, where does responsibility for AI governance sit in your org? You know, is is is it security? Is it IT? Is it legal? Is it all three? And, additionally, like, how do you avoid anything falling through the cracks? Yeah. I don't think it's one. It's... I I see it as a joint business ownership and responsibility based on where the company needs to go. I think for us as a as a dev tools company and, fortunately, or unfortunately, we find ourselves in the mix of, like, our customers want to use every model out there and the space is competitive, so we have to support that and build our way to support that. But we also have, like, very, you know, large enterprise customers who have a very different risk threshold from your average developer trying to build a new start up. Right? So how do we think about, you know, separating those environments and processes? So for us, the the AI governance is basically a shared company strategy between us and engineering and legal, and even our exec team. And we shape up what experience we want for our customers, what experience we want for our employees, and how do we allow them to have both the innovation and experimentation, but also safeguards, you know, for for our enterprise customers. And I can, you know, explain how we actually implement them too. But I think if we don't have the business buy in, on the NorthStar, then security can't drive it by itself. Thank you. So moving in kinda into our into our next section of, like, you know, what is what is good look like going into the next year? John, if you were advising a CSO starting from zero on AI governance today, what would be the first thing you would probably tell them to look at or to fix, to get ahead of it? Well, I think the first thing, goes back to a comment I made earlier. It's about... And when we were talking about visibility, it is, do do you have any visibility? And most of them don't realize that they already have some of those capabilities in place. So it's talking them through how to gather data from what they have and be able to create a bit of telemetry of data so they can understand how these dots are being connected, what's being used. But then, to the point that I just made about, you know, how to... You know, work... Working with the business, it is you you do have to have buy in because it is a collective effort between the business, between legal, between, if you've got separate compliance and privacy, groups, security. It is, all of these groups have to have a role in that. And if they don't have a seat on whatever annual governance, entity in the organization there is, they should have. But help them with how to make the business case for why they need to be there. And that they're not there to try to prevent the business. They are there to think about because you think about the the CSO role, and we've evolved, I call it over the last four generations, ten years being a generation, if you think about it, is, you know, the first generation was about being technical, but, you know, Steve Katz, and it was a technical problem because, you know, city got hacked and they were trying to solve that. They were looking at it purely technical. Then it became, okay. You've got to be aligned with the business. You know? And then the third generation was more about, okay, being risk managers. Now we have to be more about supporting and enabling the digital era, you know, which AI is, you know, leading the way in that. It is how do you become a a partner in that? How do you become an enabler? Because the security capabilities that we are looking to deploy and implement aren't to be disrupted. They are to be enabled. And so it is... I'd advise them to establish a strategy for how they enable the responsible use and secure adoption of AI. And that looks, it... It's... Has multiple parts to it. But, you know, you have your detective and protected capabilities today. But now how can you put some of those preventative capabilities in place? And I think that's where, you know, solutions like Vercel and some of the others come into play to be able to assist and support in that enable... Enabling that responsible use and secure adoption. Your roles and responsibilities are definitely changing drastically over the years, so so I definitely agree with that. Tala, where do you expect regulation or frameworks such as, like, SOC two or ISO, and this to catch up with AI risk in the next year? Yeah. I think there are already risk frameworks out there. Nest has one. EJU has one. ISO has one. There are bunch of frameworks and standards which already got launched a few years ago and are iterating. I think the way I think about this is, like, first, gotta align on your company's strategy and risk appetite of what you're trying to do. Think every company will go through this iteration of they have to adopt AI and improve their processes. So there's an internal option. And then it's also a question of the AI use and enablement in their products and services. So that's another angle they have to think too. And the kind of regulation and requirements that would apply would be depending on the type of vertical and space that, company has product and services for, whether it's highly regulated like health care or banking, different sort of requirements than if you are just, you know, selling technology software as an example. So I think that's... Again, going back to, like, it needs to line up with the company's strategy and risk appetite, and then it's easier to align business stakeholders on the level of, compliance and security and risk management that you have to apply due to their option. I do think I think there... There's there's some open ended themes that are emerging. One, obviously, is this race to win the AI, and that's... Has some geopolitical and national interests. I think the data sovereignty is definitely gaining steam for companies who want to control their data. I think you could believe various words here, which is either companies who are building frontier models, when there is or this becomes more open source and open weighty. And, but regardless of that, I think companies have to think about the mode is their data and how do they protect their data, and that's definitely an area that I expect will get more and more scrutiny in... Both in terms of, national security as well as intellectual property. And those areas are kind of, like, I think, I wouldn't say open ended, but, like, unsolved right now where I expect there will be more, regulatory pressures. Makes total sense. And, Adrian, kinda rounded us out. What's... I feel like you might even said this earlier. Like, what's one prediction about AI and security you feel confident about and one that you're kinda genuinely uncertain of? I definitely think anything I could say right now, six months later, I'll look back and I'll say, It will either have been trite and so obviously true or it'll have been so obviously wrong, but I don't know. I think the moat being data, I think, I I completely buy that. I think there was a brief while where there was a concern that the foundation models were basically gonna eat everything, that all of SAS was gonna go away. And I think that was only, like, five months ago, which is a crazy timeline to go from, oh my gosh, software is dead. It's gonna be completely eaten by these new things to, actually, hold on. Data is the center of value and the center of relevance, and AI is gonna be an enabler for companies that have SaaS products and data to be able to unlock incredibly new capabilities, which is different from the capabilities that they've built are gonna become obsolete, which was what we were thinking six months ago. So I think that's that's probably the number one thing is it's it's very... It's all gonna change. I do think, at a macro level, many of the security primitives that we have, what do good boundaries look like? What do good cryptologics look like? What is, a meaningful security model? For better or for worse, those are some of the most well established, well supported, sort of theoretical practices in software. Right? Like we know that cryptography works and isn't, you know, backed up by an NP complete problem, and we know that that's hard, and we know that factoring is relatively hard, etcetera. What we've not done a good job of is making sure that those are actually what's implemented as the security model in most software, and certainly not at an enterprise level. So I think the biggest shifts are gonna be how identity is managed, how security models, at a software level get more robust, and how those then have to be implemented at an enterprise level. We are looking very carefully at every single application that we have in our environment and trying to understand, like, what are the security boundaries? Who's allowed to access it? When is it automated? When is it not automated? You know, what how often are we doing key rotation? Can we do key rotation and secrets rotation for every single service every ten minutes? If not, why not? What would it take to get to that? So I think that process of just like grinding through the entire enterprise and getting to a point where basic security principles are upheld, I think that's gonna be the biggest shift over the next year. And I think the back end of that is actually cybersecurity is is real, in a way that it has never been in the past. Wow. Alright. So kind of rounding and closing us out here, you know, I'm gonna give the same question to all of you, quick fire. You know, what's what's what's a practical... One practical thing that every security leader should do this week? You know, what are your organizations doing to get better to enable, enable AI governance for your customers? You know, Tahoe, let's start with you. I would say go use AI, make an app on an agent, go through that experience, and you'll learn a lot by doing it yourself on what are the security capabilities and controls you currently have that will work and what is missing. Can never go wrong with a hands on keyboard. John, how about you? Yeah. I think it's, kind of, aligned with that. It is establish a preventative set of capabilities. If you're using AI to do coding in your environment, understand that you have a trusted source for, where where where the repositories are that it's getting that code from so that you can, manage the risk of any any type of, I call it a technology supply chain, attack being... Getting ahead of that because of whether whether you are, you you know, infrastructure with, container images or whether you're building an application and, you know, you're pulling from, you know, repositories, whether it's, you know, Java, Python, whatever it may be, whether you're building a business app or whether you're building some app for, some internal technology use, understanding where those snippets of code and where those components are coming from. And that's what a preventative capability is all about in in this rate... In this case. Thanks, Sharon. Adrian, round this out. I I think both of those are right. Knowing what the tools are, knowing the defensive posture and how you're trying to establish it, that... Those totally make sense. For me, what's been really instructive is every time someone asks me a security question, about the way my enterprise operates, I ask, what would it take to get to a point where I would be able to have a model plus a skill plus some set of data make this decision for me? And and and how would I have to do that? And it it could be everything. Right? It could be, a person accessed a piece of data. Was that correct? It's like, well, how would I get the right information to a model to be able to answer that question? What would it need to have access to and what would it take for me to be confident that that model was going to be operating correctly in making that decision? And and, you know, my job as CISO is basically to figure out how to do that, how to get to the point where we can express the expected outcome, the expected policy that would be applied, and then also put the systems in place so that it could do that. And for so long, we've relied on people, to sort of figure that out dynamically and track it all down dynamically. But if I could just get to the point where a tool was doing it, that would be that would be sort of Nirvana. Yeah. And and so that's that's what we're trying to build, both as a product but also for our own internal enterprise, is, like, how do you get all those things connected? How do you tie it into Rippling AI or whatever AI you're using to make those decisions, and then go from there? That's a great note to end on. Introspection, I guess, is is what I would ask for. It's a great note to end on. Adrian, John, Todd, you know, thank you for making time for this and for just being candid about a problem that I think most IT leaders and all of us are still trying to figure out in real time. So thank you for your time, and thank you for sharing. Thank you very much, mate. Yeah. It was fun. Thanks. Pleasure. Alright. Well, thanks, everyone. Hope you enjoyed that. Just wanted to go over a couple of slides and just remind about the end of the session survey. If you had a chance, go check that out. But hopefully, you enjoyed today's session. You know, we we covered a lot, but I think at the heart of it, you know, we really just wanted to get down to what does AI governance look like in the modern age. Right? And the goal with Rippling being built on what we call the employee graph here, you know, we're able to break down AI usage and spend, you know, govern tools and model access and manage agent identity. And we're also able to detect shadow AI. So, you know, as you're thinking about scaling AI in your organization, hopefully, you're keeping these these thought things in your in your head about how to kinda approach them. And that's just something that we do here. And then also just to kind of build on that, like, what does Rippling AI mean for us and and hopefully, you know, you all as a takeaway is that, you know, we're really trying to build something that is deeply ingrained in your platform and and has access to, you know, your your customer... I mean, your employee data. So, you know, what what what can we do with that? Right? Well, we can enforce and automate security policies. We can monitor and analyze your entire IT environment, and then also take action, you know, across devices, access, and systems. And I know that people, you know, when they kinda think about, well, what does take action mean? The thought process when we built Rippling AI is that you're able to, create policies based on actionable things. So the example I always like to give people is you know that someone needs to have access to Salesforce. They meet a certain piece of criteria based on data that your system already knows about. You know, why does that have to become a ticket, for someone to go action. Right? Or you know that someone wants to put in a ticket about their device running slow. You know, you can set up a policy that says if it's more than six years old and, you know, it has a certain processor, certain amount of RAM, automatically approve they... A replacement device. And that's really where we kinda centered it all around, our Rippling help desk. It's a help desk agent that can kinda help take action, based on policies that you create, to kinda cut down on those repetitive annoying tickets, that I think all of us in our IT career, you know, don't wanna have to deal with from time to time. So if you're interested in any of that, check out the doc section that we have. Feel free to reach out to us. I also just wanted to selfishly plug. We have another webinar coming up on October 28, same time, 3PM eastern, 12PM pacific. We'll be talking to some... Our product lead, AJ Yon, that oversees our automated compliance product and a couple other individuals. That's not... That's one you're definitely not gonna wanna miss. So, yeah. And then lastly, just to kind of round it out, I did wanna talk about our sweepstakes winner. So we did our we did our raffle, and today's winner is Joe Sheffield. So, Joe, if you're on, you know, congratulations for the $500 Airbnb gift card. We'll make sure to send that out to you after the webinar. But just wanted to thank everyone for their time today. Hopefully, you all found this very valuable. So, yeah, thank you, everyone, and we look forward to hearing from you.