Black Hat USA 2026 | Levent Besik, SailPoint

Clips
More from Black Hat USA 2026

Levent Besik

Chief Product Officer

SailPoint

play_circle_outline Challenges of traditional human-centric identity models with autonomous AI agents

play_circle_outline Three-Layer Agentic Fabric: Prompt, Planning, Runtime Authorization; Intent/PII Detection; Policy-Based Tool/API Controls

play_circle_outline Runtime controls: just-in-time provisioning, continuous monitoring, risk scoring

play_circle_outline Human Oversight of Agent Permissions and Credential Management to Prevent Lateral Movement

Info

Levent Besik, SailPoint

Levent Besik

Chief Product Officer SailPoint

search

Krista Case

>> Welcome back to theCUBE. This is our ongoing live coverage of day 2 here at Black Hat 2026 at the Mandalay Bay in Las Vegas. We've been talking so much over the last couple days about AI agents and how they're starting to become operational and actually interact with day-to-day business processes. Now, over these next few minutes, we're actually gonna double-click into identity. Because identity is really the mechanism for ensuring that AI agents can access and have the appropriate permissions to remain aligned with business intent. So identity is a mechanism that we're not only governing the access, but also what the agents are authorized to do. So here to join me in that conversation is Levent Besik, Chief Product Officer with SailPoint. Levent, thank you so much for joining us today.

Levent Besik

>> Thanks for having me. Great to be here.

Krista Case

>> Absolutely. So, Levent, let's start with traditional identity models. Where do we see them start to break down in this world of AI agents?

Levent Besik

>> Yeah, so the traditional human-centric identity models happen to be more static and more admin-defined policies that are basically more like batch processed. And of course, in the real-time world, that just breaks down very quickly. You know, in the agentic world, having static access privileges to a whole bunch of resources that our agents can actually exploit. And on top of that, that very quickly becomes a major liability. So I think that's one place that it breaks. The other part of it is pure proliferation of these agents, right? The average enterprise now it's 1:50 to 1:100 non-human to human identity ratio. And that immense scale makes it really hard to manage them. And it really breaks the traditional, like, the humans that join an organization, move to an organization, maybe another one, and then leave. That whole management concept kind of breaks down.

Krista Case

>> Yeah, yeah, certainly, Levent. And I know the governance component becomes a key challenge in that conversation because it's very difficult for organizations to understand really what that identity is touching and what it's doing, for lack of a better way to put it. So how should organizations think about evolving their identity governance implementations to be able to support this dynamic and autonomous agent behavior that to your earlier point is happening at speed and scale that we've never seen before?

Levent Besik

>> YeahSo first of all, I think the first thing to notice is a lot of companies are seeing this challenge on their hands and really trying to think about, hey, look, can I get a point solution in place to fix this agentic security issue? And the reality is, agentic security is deeply tied into human governance up to a certain extent. As a matter of fact, especially in this current world where many of the agents run on behalf of a user, the human security posture becomes an exposure area and angle, right, for agents. So the first thing is, how do you bring the human and data context that you have in your enterprise really tightly into your agentic security solution? That's one thing we deeply focused on in our SailPoint Agentic Fabric. Really want to make sure that it's not just a point solution for agents, but also bringing all that deep data and human context. And this is important because if you think about an agent, maybe a coding agent, let's say that, right, that is running on behalf of an engineer, right? Maybe checking in code. That's totally fine. That's low risk, right? But if you see a coding agent running on behalf of our CMO or marketing manager changing code, now that's high-risk activity, right? So that human and checking in code to where? To GitHub, maybe to a marketing repository. That context becomes super important. So I think that's really the first piece, bringing that deep identity context into an agentic solution. The second piece is, look, there is a lot of pressure in today's enterprise to really deploy an agentic anything and really reap the benefits of that. And with that, you get a lot of pressure to get them deployed very fast. In some cases without security, proper security implications and analysis and protections in place. As a matter of fact, what we find is only half of the customers that deploy agents have policies to protect them in place, right? So there's a lot of pressure to get it out there very quickly. And also end users, employees are using, bringing their own shadow AI, if you will, right, into the picture. And that becomes a major challenge for security professionals, right? And so to really, I think, address that, one key thing is getting really complete visibility.

Krista Case

>> Yep.

Levent Besik

>> Right? How do you make sure that you get full visibility into your shadow AI running on browsers, AI agents running on endpoints, all your platform agents, all in one place, end to end? And then that's the first step because if you don't see them, you cannot secure them. And seeing them is the first step to security.

Krista Case

>> I agree, Levent, and I'd be interested to get your take on this. I'm hearing that practitioners are kind of almost having to try to solve the two sides of that coin at the same time. They're trying to, to your point, get their arms around the visibility component, address shadow AI as you mentioned, but then they're also trying to put these guardrails in place to make sure that as they're uncovering these AI agents, that they're also not getting in the way of, you know, the business's ability to harness the potential for AI and implement governance. So I'd just be curious to get your take if you're seeing that from customers as well.

Levent Besik

>> 100%. The modern-day CISO right now, you know, they are not running a department of no, they're running a department of how do I enable this great technology securely? So that is the new, you know, CISO mission. And so absolutely, we definitely see that change. So it's really, and that's why when we build these protections in place, it's really about enabling and really empowering both the users and the companies to harness the power versus really stopping it. But there are really great technologies in place to do that. For instance, I'll give you an example. Agent behavioral monitoring is an area that we invested quite significantly. What this basically is, it kind of monitors an agent from the moment that it starts, starting from really the prompts that it got, the intent of the user, to the mission that it got and monitoring its activity over time, like what resources it's accessing over time. And if it ever goes outbound and maybe accesses a new resource that doesn't make sense, for instance, a marketing agent, maybe a social media agent suddenly accessing financial databases, now that's a high-risk activity. So being able to just have that policy in place to say, you know what, over a certain risk or risk score, pause this agent or maybe disable it until my SOC team can look into it. I think that's the kind of technology that we're putting in place to enable this versus starting from a department of no sense.

Krista Case

>> Yes, I love that kind of becoming a department of yes. And Levent, I really appreciate that you're talking about this is all grounded in the risk level because we are seeing that, of course, security teams, they can't address everything. They're being inundated. So they really need to prioritize what's critical to the business. So again, just something that jumped out there. So I wanted to also talk a little bit more about the runtime conversation that we've been having and going back to this conversation around AI speed and kind of becoming a department of yes and not getting in the way, one challenge that I'm hearing when I talk to practitioners is as we try to govern AI agents, there's not enough time for the human to approve a privileged access request, for example. So, can you talk about the authorization and governance and how that needs to evolve in this agentic AI world?

Levent Besik

>> Absolutely. Yeah. And we approach this in a layered way because we think that there's defense in depth that you need to deploy here. In our Agentic Fabric, we have effectively 3 layers of protections that we provide. First layer really starts at the prompt, right? At the prompt layer, what is the intent of the prompt? What is— and does the prompt include any kind of poisonous instructions in there? Is there any PII in there or any confidential data being uploaded? That it's protecting both the agent and the human user interaction. So the user itself, right, from potential challenges, right? So it really starts there, but getting the intent is a really critical piece there. And so our fabric evaluates it with policies. And then once it passes that, the second layer becomes, of course, the agentic planning with the MCP tools and resources exposed or the API access. At that level, our Agentic Fabric also plugs in at every level and we meet the customers wherever they are. They might have local MCP servers, they might have an MCP gateway, right? Or we just plug into their platforms. But as they plan, as the agent plans, what do they have or should they have access to those resources or what should they see even? What should their world include or not include, right? So that's a second layer of protection. And then on the third layer, when actually agent decides to go ahead and take action on that plan, we have additional layers. One is, SailPoint is a world leader in provisioning, deprovisioning access to all kinds of identities, including non-human identities. We have over 5 billion entitlements provisioned through SailPoint platform. So we can actually just in time, in real time, provision entitlements or privileges either on behalf of the user or on behalf of the agent. So that, access is only based on policy,

Krista Case

>> right?Which is especially— I know standing privilege has been a challenge for many years, but it's especially a challenge with the AI agents. Not to interrupt you,

Levent Besik

>> but—No, absolutely right. And honestly, this is also maybe we can come back to it, but the human security posture augments and informs the agentic security posture as well. Because, if you have standing privileges to, privilege, financial information, suddenly your agent has access to that, right? There's no control over it, right? Unless you have the policies in place. So our Just-in-time provisioning solves that problem. And we offer that for Identity Fabric too. So we provision access. We obviously work with the IDP on your authorization policies, right? Whatever needs to happen. And at that third layer, now the agent is executing whatever action it's trying to execute. Our job is not done, right? Because now we have to continuously monitor agent activity. What if it steps out of bounds? What if it, drifts? What if, it does something like what happened with Hugging Face, right?

Krista Case

>> Yes.

Levent Besik

>> This is where, having those policies is super important, the agent behavioral policies where we constantly monitor risk, we assign a risk score to every agent. And if the risk is above a certain threshold, you can either continuously, monitor and block or disable it. Or

Krista Case

>> maybe—the intent as well becomes important in that conversation.

Levent Besik

>> Absolutely right. Or maybe we get a, a Shared Signals Framework signal that identity was compromised, the device was compromised. So, we also respond to that in real time. So I think this is critical, right? Being able to protect and respond is really critical in this new world. And those are kind of the 3 layers, I would say, right? Both the prompt security, the planning stage, and the runtime authorization and continuous authorization.

Krista Case

>> So Levent, I'm really interested to get your take on kind of the role that this plays in building that trust as these AI agents become part of day-to-day processes. I mean, the narrative that I hear is kind of it’s what really facilitates the secure access to these data and resources. But if you could elaborate, your perspective on that.

Levent Besik

>> Yeah, I mean, the data access is critical, right? 'Cause let's be honest, everything starts with data and ends with data, right? So in our world, what we pay a lot of attention to is making sure that we bring that data context deeply into all kinds of policies, right, that you can run. At authorization layer or maybe even access layer or provisioning layer. We integrate deeply into all your top data providers out there and we become a critical piece of that equation, right? By protecting that access for, 'cause, again, the agentic access to data is a critical path that we need to protect and definitely our Agentic Fabric helps with that.

Krista Case

>> Yep, absolutely. And so here at Black Hat, not only in the identity space, but in every space we're seeing a slew of AI-driven capabilities. I'm curious what do you think practitioners are going to expect from their identity platforms in terms of some of these AI-driven capabilities maybe over the next 12 to 18 months as they're trying to navigate this new world?

Levent Besik

>> Yeah, it's been incredible here this year, right? there's so much energy and, you know, obviously we had recent incidents that really suddenly put Security in the forefront, right?

Krista Case

>> You mentioned Hugging Face, and it's a great example of this very problem, what can potentially happen when these AI agents are not appropriately governed.

Levent Besik

>> Yeah, absolutely. Look, when I talk to CISOs and security practitioners, the simple, I would say, framework that we offer with a lot of obviously great capabilities are really threefold: discovering agents, end-to-end wherever they might be, right? That's the first step. And then really being able to govern them, lifecycle manage them, making sure that they have human owners. if the human owner leaves, there is another human owner because you want accountability in place, making sure all the agents' activity gets logged, and then being able to audit them properly for your compliance purposes and compliance rituals. And then the third step being protecting and authorizing them, right, properly in real time. So, I think discovery is critical because as I said, you cannot secure what you don't see.

Krista Case

>> Exactly. Exactly.

Levent Besik

>> And one thing that we pay a lot of attention to at SailPoint is how do we make sure that we have the complete unified inventory of agents. We have now very lightweight browser sensors and endpoint sensors that sense and enforce policy on both browser agent interactions for the user or on the endpoint with Claude Code and everything else. Right. And of course, we plug into all the platforms. So we're bringing all this agentic activity and all the types of agents and also non-human identities associated with those agents.

Krista Case

>> Okay.

Levent Besik

>> Because if you look at Hugging Face, for instance, the models that escaped, you know, found exposed credentials and kind of laterally move, right? So exposed credentials become super important to make sure that you handle. So again, not a new problem, But definitely a huge problem now because agents can quickly get hold of those and exploit them. Right. So getting access to, again, a complete inventory of your non-human identities, those credentials, and understanding your security posture really across both human and non-human and making sure your standing privileges are closed, right, to just-in-time access, making sure there's nothing out there standing because that's an invitation for exploitation as well. Right. And that's really the first step, right? Getting that full inventory. And once you have that inventory, then enforcing human ownership, making sure that every agent has a human owner and it's very clear who that is, being able to assign policies around them, right? With birthright permissions, like what the agents are allowed and not allowed to do the moment that we find them or they pop up in your digital landscape, right? And what happens if the owner moves? Those kinds of lifecycle policies are critical. So that's basically governance in our world, right? And then we talked about the last piece quite a bit, right? Which is these 3 layers of authorization and protections that we offer, right? From prompt security to real-time authorization, continuous authorization. And there's one more piece, those 3 will give you an excellent, agentic security posture and deploying those with our Fabric. But if incidents happen, what we also see is agentic identity and all the identity context is also critical for investigations for your security operation center. So being able to bring that identity context into SOAR, into SOC is critical. And actually that's another piece that we announced here at Black Hat today. I should say yesterday that we now also have a SailPoint Agentic and Human Identity Intelligence package being exposed and integrated into top platforms like CrowdStrike, right? So the security professionals and responders can immediately see all that identity context. So I think with all that discovery, governance, and protection on top of that, you know, identity context and response, It's kind of the recipe that we kind of offer to our customers.

Krista Case

>> Yes, that makes sense. Levent, a couple of reactions. First, I do like that you're kind of marrying the human with the AI agent. I think we've been spending a lot of time thinking about how agents are different than humans, and we're kind of maybe overlookingthe correlations that we need to draw to your point of, okay, this agent is going to potentially inherit the permissions of the human. So I really like that you're kind of marrying that context. And then on the discovery piece, I know you and I got the opportunity to sit down a couple of days ago and you were talking to some customers that maybe have 100x more AI agents in their environment than they thought. Right. So I think that's kind of addressing that visibility piece is going to be very critical.

Levent Besik

>> You know, it's fascinating because with these tools now, and by the way, maybe I should talk about this a little bit too, like what surprises CISOs the most when they deploy some of these tools. What we see is, first of all, on our side, a pleasant surprise is how quickly we can get deployed. And within hours show that value. Literally, you can deploy our Fabric and get agentic posture information, like how many agents you have, how many secrets you have on both non-human side and agentic side, I should say, very quickly within hours. So that's a pleasant surprise because that's not what they're used to from the old human world, right? But the one that I think shocks them is, as you said, the immense number. The numbers of agents that we discover are usually 10x to 100x more. we had situations where a security professional said, oh yeah, I have no agents, everything's locked down. And we said, no, no, you actually have 1,000 agents, right? Some of them with no human owners, you don't see them, they're shadow AI, right? It's happening for you. So that's a big surprise. The other thing now, of course, is I kind of mentioned this earlier, right? All these other posture pieces like exposed secrets or credentials, right? floating around and unencrypted in places not in vault, unmanaged credentials, right? That becomes a threat vector. So in a lot of these POCs, we see our customers are surprised by us finding a large number of, for instance, exposed credentials waiting to be exploited, right? So again, those are some pleasant and some unpleasant but important realizations that security professionals see. And you're right, these numbers I think are only going to increase. As agent proliferation happens.

Krista Case

>> Well, even if it's a little bit uncomfortable, it's so critical to have that visibility because then it can be addressed.

Krista Case

>> So that's right.

Krista Case

>> Absolutely. Well, Levent, thank you so much for sitting down with us here at theCUBE today. We really appreciate it.

Levent Besik

>> Oh, thank you for having me. It was great fun.

Krista Case

>> Thank you so much. And thanks for joining. Don't go away. We're going to be right back in just a couple minutes with more from Black Hat 2026.

Levent Besik, SailPoint

search

Krista Case

>> Welcome back to theCUBE. This is our ongoing live coverage of day 2 here at Black Hat 2026 at the Mandalay Bay in Las Vegas. We've been talking so much over the last couple days about AI agents and how they're starting to become operational and actually interact with day-to-day business processes. Now, over these next few minutes, we're actually gonna double-click into identity. Because identity is really the mechanism for ensuring that AI agents can access and have the appropriate permissions to remain aligned with business intent. So identity is a mechanism that we're not only governing the access, but also what the agents are authorized to do. So here to join me in that conversation is Levent Besik, Chief Product Officer with SailPoint. Levent, thank you so much for joining us today.

Levent Besik

>> Thanks for having me. Great to be here.

Krista Case

>> Absolutely. So, Levent, let's start with traditional identity models. Where do we see them start to break down in this world of AI agents?

Levent Besik

>> Yeah, so the traditional human-centric identity models happen to be more static and more admin-defined policies that are basically more like batch processed. And of course, in the real-time world, that just breaks down very quickly. You know, in the agentic world, having static access privileges to a whole bunch of resources that our agents can actually exploit. And on top of that, that very quickly becomes a major liability. So I think that's one place that it breaks. The other part of it is pure proliferation of these agents, right? The average enterprise now it's 1:50 to 1:100 non-human to human identity ratio. And that immense scale makes it really hard to manage them. And it really breaks the traditional, like, the humans that join an organization, move to an organization, maybe another one, and then leave. That whole management concept kind of breaks down.

Krista Case

>> Yeah, yeah, certainly, Levent. And I know the governance component becomes a key challenge in that conversation because it's very difficult for organizations to understand really what that identity is touching and what it's doing, for lack of a better way to put it. So how should organizations think about evolving their identity governance implementations to be able to support this dynamic and autonomous agent behavior that to your earlier point is happening at speed and scale that we've never seen before?

Levent Besik

>> YeahSo first of all, I think the first thing to notice is a lot of companies are seeing this challenge on their hands and really trying to think about, hey, look, can I get a point solution in place to fix this agentic security issue? And the reality is, agentic security is deeply tied into human governance up to a certain extent. As a matter of fact, especially in this current world where many of the agents run on behalf of a user, the human security posture becomes an exposure area and angle, right, for agents. So the first thing is, how do you bring the human and data context that you have in your enterprise really tightly into your agentic security solution? That's one thing we deeply focused on in our SailPoint Agentic Fabric. Really want to make sure that it's not just a point solution for agents, but also bringing all that deep data and human context. And this is important because if you think about an agent, maybe a coding agent, let's say that, right, that is running on behalf of an engineer, right? Maybe checking in code. That's totally fine. That's low risk, right? But if you see a coding agent running on behalf of our CMO or marketing manager changing code, now that's high-risk activity, right? So that human and checking in code to where? To GitHub, maybe to a marketing repository. That context becomes super important. So I think that's really the first piece, bringing that deep identity context into an agentic solution. The second piece is, look, there is a lot of pressure in today's enterprise to really deploy an agentic anything and really reap the benefits of that. And with that, you get a lot of pressure to get them deployed very fast. In some cases without security, proper security implications and analysis and protections in place. As a matter of fact, what we find is only half of the customers that deploy agents have policies to protect them in place, right? So there's a lot of pressure to get it out there very quickly. And also end users, employees are using, bringing their own shadow AI, if you will, right, into the picture. And that becomes a major challenge for security professionals, right? And so to really, I think, address that, one key thing is getting really complete visibility.

Krista Case

>> Yep.

Levent Besik

>> Right? How do you make sure that you get full visibility into your shadow AI running on browsers, AI agents running on endpoints, all your platform agents, all in one place, end to end? And then that's the first step because if you don't see them, you cannot secure them. And seeing them is the first step to security.

Krista Case

>> I agree, Levent, and I'd be interested to get your take on this. I'm hearing that practitioners are kind of almost having to try to solve the two sides of that coin at the same time. They're trying to, to your point, get their arms around the visibility component, address shadow AI as you mentioned, but then they're also trying to put these guardrails in place to make sure that as they're uncovering these AI agents, that they're also not getting in the way of, you know, the business's ability to harness the potential for AI and implement governance. So I'd just be curious to get your take if you're seeing that from customers as well.

Levent Besik

>> 100%. The modern-day CISO right now, you know, they are not running a department of no, they're running a department of how do I enable this great technology securely? So that is the new, you know, CISO mission. And so absolutely, we definitely see that change. So it's really, and that's why when we build these protections in place, it's really about enabling and really empowering both the users and the companies to harness the power versus really stopping it. But there are really great technologies in place to do that. For instance, I'll give you an example. Agent behavioral monitoring is an area that we invested quite significantly. What this basically is, it kind of monitors an agent from the moment that it starts, starting from really the prompts that it got, the intent of the user, to the mission that it got and monitoring its activity over time, like what resources it's accessing over time. And if it ever goes outbound and maybe accesses a new resource that doesn't make sense, for instance, a marketing agent, maybe a social media agent suddenly accessing financial databases, now that's a high-risk activity. So being able to just have that policy in place to say, you know what, over a certain risk or risk score, pause this agent or maybe disable it until my SOC team can look into it. I think that's the kind of technology that we're putting in place to enable this versus starting from a department of no sense.

Krista Case

>> Yes, I love that kind of becoming a department of yes. And Levent, I really appreciate that you're talking about this is all grounded in the risk level because we are seeing that, of course, security teams, they can't address everything. They're being inundated. So they really need to prioritize what's critical to the business. So again, just something that jumped out there. So I wanted to also talk a little bit more about the runtime conversation that we've been having and going back to this conversation around AI speed and kind of becoming a department of yes and not getting in the way, one challenge that I'm hearing when I talk to practitioners is as we try to govern AI agents, there's not enough time for the human to approve a privileged access request, for example. So, can you talk about the authorization and governance and how that needs to evolve in this agentic AI world?

Levent Besik

>> Absolutely. Yeah. And we approach this in a layered way because we think that there's defense in depth that you need to deploy here. In our Agentic Fabric, we have effectively 3 layers of protections that we provide. First layer really starts at the prompt, right? At the prompt layer, what is the intent of the prompt? What is— and does the prompt include any kind of poisonous instructions in there? Is there any PII in there or any confidential data being uploaded? That it's protecting both the agent and the human user interaction. So the user itself, right, from potential challenges, right? So it really starts there, but getting the intent is a really critical piece there. And so our fabric evaluates it with policies. And then once it passes that, the second layer becomes, of course, the agentic planning with the MCP tools and resources exposed or the API access. At that level, our Agentic Fabric also plugs in at every level and we meet the customers wherever they are. They might have local MCP servers, they might have an MCP gateway, right? Or we just plug into their platforms. But as they plan, as the agent plans, what do they have or should they have access to those resources or what should they see even? What should their world include or not include, right? So that's a second layer of protection. And then on the third layer, when actually agent decides to go ahead and take action on that plan, we have additional layers. One is, SailPoint is a world leader in provisioning, deprovisioning access to all kinds of identities, including non-human identities. We have over 5 billion entitlements provisioned through SailPoint platform. So we can actually just in time, in real time, provision entitlements or privileges either on behalf of the user or on behalf of the agent. So that, access is only based on policy,

Krista Case

>> right?Which is especially— I know standing privilege has been a challenge for many years, but it's especially a challenge with the AI agents. Not to interrupt you,

Levent Besik

>> but—No, absolutely right. And honestly, this is also maybe we can come back to it, but the human security posture augments and informs the agentic security posture as well. Because, if you have standing privileges to, privilege, financial information, suddenly your agent has access to that, right? There's no control over it, right? Unless you have the policies in place. So our Just-in-time provisioning solves that problem. And we offer that for Identity Fabric too. So we provision access. We obviously work with the IDP on your authorization policies, right? Whatever needs to happen. And at that third layer, now the agent is executing whatever action it's trying to execute. Our job is not done, right? Because now we have to continuously monitor agent activity. What if it steps out of bounds? What if it, drifts? What if, it does something like what happened with Hugging Face, right?

Krista Case

>> Yes.

Levent Besik

>> This is where, having those policies is super important, the agent behavioral policies where we constantly monitor risk, we assign a risk score to every agent. And if the risk is above a certain threshold, you can either continuously, monitor and block or disable it. Or

Krista Case

>> maybe—the intent as well becomes important in that conversation.

Levent Besik

>> Absolutely right. Or maybe we get a, a Shared Signals Framework signal that identity was compromised, the device was compromised. So, we also respond to that in real time. So I think this is critical, right? Being able to protect and respond is really critical in this new world. And those are kind of the 3 layers, I would say, right? Both the prompt security, the planning stage, and the runtime authorization and continuous authorization.

Krista Case

>> So Levent, I'm really interested to get your take on kind of the role that this plays in building that trust as these AI agents become part of day-to-day processes. I mean, the narrative that I hear is kind of it’s what really facilitates the secure access to these data and resources. But if you could elaborate, your perspective on that.

Levent Besik

>> Yeah, I mean, the data access is critical, right? 'Cause let's be honest, everything starts with data and ends with data, right? So in our world, what we pay a lot of attention to is making sure that we bring that data context deeply into all kinds of policies, right, that you can run. At authorization layer or maybe even access layer or provisioning layer. We integrate deeply into all your top data providers out there and we become a critical piece of that equation, right? By protecting that access for, 'cause, again, the agentic access to data is a critical path that we need to protect and definitely our Agentic Fabric helps with that.

Krista Case

>> Yep, absolutely. And so here at Black Hat, not only in the identity space, but in every space we're seeing a slew of AI-driven capabilities. I'm curious what do you think practitioners are going to expect from their identity platforms in terms of some of these AI-driven capabilities maybe over the next 12 to 18 months as they're trying to navigate this new world?

Levent Besik

>> Yeah, it's been incredible here this year, right? there's so much energy and, you know, obviously we had recent incidents that really suddenly put Security in the forefront, right?

Krista Case

>> You mentioned Hugging Face, and it's a great example of this very problem, what can potentially happen when these AI agents are not appropriately governed.

Levent Besik

>> Yeah, absolutely. Look, when I talk to CISOs and security practitioners, the simple, I would say, framework that we offer with a lot of obviously great capabilities are really threefold: discovering agents, end-to-end wherever they might be, right? That's the first step. And then really being able to govern them, lifecycle manage them, making sure that they have human owners. if the human owner leaves, there is another human owner because you want accountability in place, making sure all the agents' activity gets logged, and then being able to audit them properly for your compliance purposes and compliance rituals. And then the third step being protecting and authorizing them, right, properly in real time. So, I think discovery is critical because as I said, you cannot secure what you don't see.

Krista Case

>> Exactly. Exactly.

Levent Besik

>> And one thing that we pay a lot of attention to at SailPoint is how do we make sure that we have the complete unified inventory of agents. We have now very lightweight browser sensors and endpoint sensors that sense and enforce policy on both browser agent interactions for the user or on the endpoint with Claude Code and everything else. Right. And of course, we plug into all the platforms. So we're bringing all this agentic activity and all the types of agents and also non-human identities associated with those agents.

Krista Case

>> Okay.

Levent Besik

>> Because if you look at Hugging Face, for instance, the models that escaped, you know, found exposed credentials and kind of laterally move, right? So exposed credentials become super important to make sure that you handle. So again, not a new problem, But definitely a huge problem now because agents can quickly get hold of those and exploit them. Right. So getting access to, again, a complete inventory of your non-human identities, those credentials, and understanding your security posture really across both human and non-human and making sure your standing privileges are closed, right, to just-in-time access, making sure there's nothing out there standing because that's an invitation for exploitation as well. Right. And that's really the first step, right? Getting that full inventory. And once you have that inventory, then enforcing human ownership, making sure that every agent has a human owner and it's very clear who that is, being able to assign policies around them, right? With birthright permissions, like what the agents are allowed and not allowed to do the moment that we find them or they pop up in your digital landscape, right? And what happens if the owner moves? Those kinds of lifecycle policies are critical. So that's basically governance in our world, right? And then we talked about the last piece quite a bit, right? Which is these 3 layers of authorization and protections that we offer, right? From prompt security to real-time authorization, continuous authorization. And there's one more piece, those 3 will give you an excellent, agentic security posture and deploying those with our Fabric. But if incidents happen, what we also see is agentic identity and all the identity context is also critical for investigations for your security operation center. So being able to bring that identity context into SOAR, into SOC is critical. And actually that's another piece that we announced here at Black Hat today. I should say yesterday that we now also have a SailPoint Agentic and Human Identity Intelligence package being exposed and integrated into top platforms like CrowdStrike, right? So the security professionals and responders can immediately see all that identity context. So I think with all that discovery, governance, and protection on top of that, you know, identity context and response, It's kind of the recipe that we kind of offer to our customers.

Krista Case

>> Yes, that makes sense. Levent, a couple of reactions. First, I do like that you're kind of marrying the human with the AI agent. I think we've been spending a lot of time thinking about how agents are different than humans, and we're kind of maybe overlookingthe correlations that we need to draw to your point of, okay, this agent is going to potentially inherit the permissions of the human. So I really like that you're kind of marrying that context. And then on the discovery piece, I know you and I got the opportunity to sit down a couple of days ago and you were talking to some customers that maybe have 100x more AI agents in their environment than they thought. Right. So I think that's kind of addressing that visibility piece is going to be very critical.

Levent Besik

>> You know, it's fascinating because with these tools now, and by the way, maybe I should talk about this a little bit too, like what surprises CISOs the most when they deploy some of these tools. What we see is, first of all, on our side, a pleasant surprise is how quickly we can get deployed. And within hours show that value. Literally, you can deploy our Fabric and get agentic posture information, like how many agents you have, how many secrets you have on both non-human side and agentic side, I should say, very quickly within hours. So that's a pleasant surprise because that's not what they're used to from the old human world, right? But the one that I think shocks them is, as you said, the immense number. The numbers of agents that we discover are usually 10x to 100x more. we had situations where a security professional said, oh yeah, I have no agents, everything's locked down. And we said, no, no, you actually have 1,000 agents, right? Some of them with no human owners, you don't see them, they're shadow AI, right? It's happening for you. So that's a big surprise. The other thing now, of course, is I kind of mentioned this earlier, right? All these other posture pieces like exposed secrets or credentials, right? floating around and unencrypted in places not in vault, unmanaged credentials, right? That becomes a threat vector. So in a lot of these POCs, we see our customers are surprised by us finding a large number of, for instance, exposed credentials waiting to be exploited, right? So again, those are some pleasant and some unpleasant but important realizations that security professionals see. And you're right, these numbers I think are only going to increase. As agent proliferation happens.

Krista Case

>> Well, even if it's a little bit uncomfortable, it's so critical to have that visibility because then it can be addressed.

Krista Case

>> So that's right.

Krista Case

>> Absolutely. Well, Levent, thank you so much for sitting down with us here at theCUBE today. We really appreciate it.

Levent Besik

>> Oh, thank you for having me. It was great fun.

Krista Case

>> Thank you so much. And thanks for joining. Don't go away. We're going to be right back in just a couple minutes with more from Black Hat 2026.