Agentic AI vs. Disconnected Apps: What It Can and Can't Solve

Hi there. I'm Tom Field. I'm senior vice president of editorial at IS and G. I wanna welcome you to our session today, where the topic is AgenTic AI versus identity's last mile problem. Your presenter is Matt Coyote. He is the CSO at Cerby And before I bring him on to our virtual stage here, let me give you some background on our session. Adjunctica AI is reshaping identity security by autonomously planning, adapting, and executing controls across the enterprise. But the last mile of identity tied to disconnected applications and manual processes remains where attackers gain their biggest advantage. Nearly half of enterprise applications still all fall outside traditional identity systems, leaving gaps managed through spreadsheets, emails, and help desk workflows. In this session today, Matt Coyote will examine real world breach scenarios and explore what AgenTeq AI can and cannot solve today. First, a little bit about my organization. ISMG is a global education and intelligence firm. We are based in the US in Princeton, New Jersey, and you may know us by any number of our thirty eight media properties. They include bank info security, data breach today, a I today dot I o. In all, we serve an audience of over two million security and technology leaders globally, and we give them a daily diet of news, analysis, research, events, and educational programs just like this one. Just a few notes of housekeeping. If you have any questions for Matt, submit them anytime by the chat window on your screen. Even if we can't get to your question today, we'll get a response back to you via email. Should you encounter technical issues while viewing today's webinar, take down that email address you see right now. If you write to webinars at I s m g dot I o, we have got support staff standing by to help. A reminder, today's webinar is copyrighted material meant for today's session and individual study purposes only. If you would like to use any of the information presented today or if you're looking for customized training materials, do contact us. I am delighted to introduce our sponsor, Cerby is the identity automation platform purpose built to secure disconnected applications, those that fall outside the reach of traditional identity security tools. By integrating with existing IAM, IGA, and PAM systems, Cerby brings centralized access controls, automates manual security risks, and extends governance across your entire application ecosystem. IT and security teams gain complete visibility and control, reducing risk and operational overhead. Founded in twenty twenty, Cerby is backed by leading investors entrusted by global enterprises, you can learn more at Cerby dot com. And now let's meet Matt Coyote. He brings more than twenty years of security leadership experience from Palo Alto Networks, RedLock, Cognizant, Deloitte, and eBay to his current role as chief strategy officer at Cerby And with that, I wanna bring on to our virtual stage, Matt Coyote. Matt, it's a pleasure to have you here. Thanks so much for being a part of this webinar today. Hey. It's great to be here, Tom. Thanks for having me. So, Matt, you have spent twenty six years as a CISO and security leader, including, of course, Palo Alto Networks. Now you're ready to start up, and you're telling people they're not ready for AI agents. What's the conversation you're actually having today with security leaders? Yeah. It's a great question. I I can't I literally cannot have a conversation without security leaders asking me about how to secure AI and AI agents. Right? And the bigger point and the reason that conversation is happening, Tom, is that every security leader is getting pressure from the business to move fast on AI agents. The challenge is that their identity stack wasn't built for this. In fact, there's new research that shows that eighty eight percent of enterprises reported an AI agent security incident in the last twelve months. So security leaders, right, they wanna know how to manage agentic identities or NHIs, and that's exactly what I wanna talk about over the next thirty or so minutes. Beautiful. Let's move forward. Alright. Now there's an image. Man, I don't wanna put some mistakes on the table here. AI agents are everywhere. Every vendor's talking about them. You and I both been to RSAC. I was at Google Next a couple weeks ago. Everyone's talking about agents. But what specifically makes this a security problem and not just a technology opportunity? I love the way you put that. I I think the challenge is is that most people are familiar now with with AI, but they think in terms of chatbots. Right? Like ChatGPT. Ask it a question, it gives you a response. The difference with an agent is that it acts autonomously on your behalf or on behalf of the organization. Right? Think of an agent as a really brilliant intern. Right? It may last con lack context, but is really brilliant. But what do they do? Right? They authenticate. They have access to data. They take tokens at machine speed and scale. The security problem is this. Right? Your identity controls in most enterprises, they were built for humans. They assume a person is logging in making deliberate decisions. The challenge is, as you can see from the screenshot, is that agents don't log in the way humans do. Right? And they touch apps that your identity stack can't see. That means you have no visibility, no control. And, you know, this specific incident we're talking about here, this was November of last year. Here, Anthropic, they documented and disrupted what they called the first large scale AI orchestrated attack. Right? In this case here, it was a Chinese state sponsored group, GTG one thousand two, and they used Claude's agentic capabilities to execute eighty to ninety percent of the attack autonomously. The AI researched the vulnerabilities, wrote the exploit code, and harvested credentials without a human directing any step. And, you know, while that was absolutely shocking at the time, right, this now happens almost every single week. And the uncomfortable part, Tom, is that the enterprise identity in most organizations is still completely manual, and this is what I call that last mile of identity that's connected to disconnected apps. And it's exactly the kind of automation that gives attackers their their biggest advantage. So what I'm hearing, Matt, the issue isn't that agents exist. It's that identity controls can only see some of what these agents are doing. And, of course, now you've shown us documented proof. Someone's already exploiting that gap. So let's take a look at what that gap actually looks like. You know, the the old movie line used to be show me the money. I'm gonna ask you. Show me the gap. Show me the gap. Alright. I'm gonna have to watch Jerry Maguire now after, after this For those that remember that. So every organization has two categories of apps. Right? Those that are connected to your identity provider and those that are disconnected. And everyone's familiar with these terms here. Right? You've got your identity stack. It things like your IGA platforms, your PAM, your enterprise password managers. Right? Organizations, enterprises specifically, they have made these investments in all these tools. The challenge is that these existing identity investments, they're great, and they work wonderful with apps that connect to them. Right? Now when we're talking about an agent, right, your agent, it touches these apps. Right? And when that is when your IDP sees that authentication event. The challenge is is when we start getting into disconnected apps that really lie below the surface of your existing identity infrastructure, that's where the disconnected apps come in. Right? And these disconnected apps are all over the place. Right? They're they're SaaS. They're on prem. They're cloud. They're legacy apps. They are all over the place. And this gap with these disconnected apps, they existed long before AI agents. Right? Agents, they didn't they didn't create this gap, but they find it and they operate in it at machine speed. Now interestingly enough, earlier this year, Panaman found in their research that eighty seven percent of organizations have already adopted AI or GenAI. And most of those AI powered apps, guess what? They don't support the technologies that the agents need in order to work. SAML for single sign on, OIDC for authentication. So that disconnected app layer is actually growing faster than most identity programs can follow it. Okay. So what I'm hearing is it's not a new gap. This existed before agents came along. You know, it's just exacerbated. It is. You're right. That's exactly right. The gap, again, that was a preexisting condition. Right? Before agents, a human might access a disconnected app. Right? It's slow. It's manual. It's constrained by human speed and attention. But now you have an agent that's, you know, accessing dozens to maybe hundreds of these disconnected apps all at once using stored credentials, and you are essentially blind from your existing identity stack perspective. So the agent, Tom, it doesn't create the vulnerability. It just makes the exposure faster and much larger, and that is exactly what we mean by the last mile of identity. Okay. So it's a structural problem you're describing is what I'm hearing. Do you have any data you can share on how widespread this actually is? Yeah. I wanna walk through some research here just to kind of orient the the followers or the listeners around the slide. So this is data that I was just talking about earlier. This was data that's very fresh just from about a month ago that was done by the Ponemon Institute with over six hundred IT and security leaders. K? There's an independent study. And, you know, just to kinda give you some orientation here. Left hand side of your screen, this is what I call the scope of the data, the scope and the scale. I'm gonna build the left and the right side out, and you'll see the right side talks really about the urgency and the trend. So let let's let's walk through this. Right? So what did the data find? Well, first off, a lot of times when I speak to security leaders about this whole concept of connected and disconnected apps, sometimes they say, well, is it really a massive part of our organization? Is it a big part of our enterprise identity stack? And, you know, it's funny. When I talk to those that are specifically in identity and access management, they don't question these. It's when you get a couple layers out where they may ask this question. Well, the data points us to the fact that about thirty percent of an enterprise's apps, they're disconnected from centralized identity systems on average. Right? The the the urgency there is that fifty eight percent of organizations say that they're that the urgency around securing these is greater than ever. Right? So think of it this way. Thirty percent of your apps. Right? So if you've got a hundred apps in your org, that means thirty of them, you can't manage centrally with your IDP. And in an organization, that's more than eighty apps that are sitting outside of your identity control plane. And if forty percent of those, right, if forty percent of those are business critical, we're not talking shadow IT. We're not talking about convenience app. We're talking core operations. And the trend line just makes it worse, Tom. Right? Fifty eight percent fifty eight percent say the number of disconnects increased in the last twelve months, and eighty seven percent, right, we know this, have adopted gen AI Gen AI. And, again, most of those AI powered apps, they don't support the identity standards that your existing identity stack uses. So every new AI tool is another disconnected app by default. And fifty six percent say the urgency to address this here has increased. So the gap, it's not getting larger. It is getting larger. It's not getting smaller. And, you know, interestingly enough, the other thing the data shows is that, you know, when we ask leaders about their confidence and their identity coverage, they only give themselves a six point three out of ten. Right? That confidence reflects the apps that they can already see. It's not that eighty plus apps that they can't. Okay. I guess sit and let that chill run through me for a minute. Matt, you wanna take a minute and and do an audience poll? Yeah. Let's do that. I think it is always good to kinda get a pulse from the audience. Let's just take let's take thirty seconds. And the question is this. Right? Do you still manage identity and access management or identity governance for at least twenty percent of your apps? Let just be honest here. Right? No one's gonna get a report card and all that. Just be honest. Do you still manage? So imagine hearing theme music as you as you respond to this poll now. That'll be interesting to see kinda as some of these results come in, Tom, here. You know? It's always interesting when we run this live. Okay. Talk about the results to and and what you what what what do results tell us? You know, you know, when when we do this, the results typically come in one of two ways. One is, like, the majority of the audience says, hey. I've I've been there. Right? That means that manual work is their exposure. Right? It means every app is managed manually or maybe that your app doesn't your IDP doesn't fully govern these apps. Sometimes when we see the data mixed, you know, that range tells the story. It just means simply that no organization has solved this. And so the question is whether you're measuring it at all. Now let's let's let's look at what happens when an agent starts touching some of those apps. Alright. So here's where we make things concrete. If an agent is running in my environment right now and it tries to access an app, what really happens? Walk me through the case. Yeah. So there really is. It's it's very binary. Right? We'll walk through both of these. There's this one here, let's talk about, as you asked, the connected app. Right? In this case here, you've got an app or an agent that authenticates to some kind of known application. Right? Maybe it's Salesforce, and it's authenticating through your IDP. Maybe you're using Okta. Maybe you're using Entra, something like that. Right? In that case, your IDP, it sees the event. Right? Conditional access policies apply, sessions can be reviewed, and you essentially have a full audit trail. Like, you are in control. This is this is the let's call this the happy path, Tom. Right? Like, your your agent, it's fully authenticated. You can see everything. You can use all the wonderful tooling and controls that are available in Entre, Okta, whatever you're using for your IDP. So this here, what we're looking at, this is the happy path. You're in full control. Now let's let's let's let's flip the flip the table, so to say. Right? Let's look at this from a disconnected app perspective. In this case here, your agent authenticates, let's just say, to a a financial reporting tool. Maybe it's an ERP, something like that. Because this is a disconnected app, right, what happens is that the agent authenticates, but it's using more than likely stored credentials. Right? It's not using anything with your IDP. Because it's not managed, the identity itself is not managed by your identity stack. Your IDP never sees it. No alerts fire. There's no policies that apply, and no one knows what's happening unless you're specifically looking. Right? So this is the same agent in the previous scenario, same intent, but there are completely different outcomes. And it's based entirely on whether your app stack your app is connected to your identity stack. That's the core problem. It's not about the agents. It's about the apps itself. Right? And, again, what are the impacts here? Right? So if your agent does something rogue, we'll talk about a specific scenario here in a minute. Maybe it's exporting data. Maybe it's changing settings. Again, there's no policy enforcement here. The agent authenticates, and it just does whatever it can do with those standing permissions. Matt, what I'm hearing is it comes down to a binary. Does your identity provider see it or not? And most organizations do not know which of their apps fall into which category. Fair? Yeah. Fair point. Fair point. Alright. So now you've just shown us the terrain. Your IDP sees it or it doesn't. Now show me what an attacker does with that. Walk me through this scenario. Yeah. I I think that it's it's really important. Most organizations again, we found that eighty eight percent, they have experienced some type of agentic based incident they've had in the last year. But what I wanna do is I wanna give a a hypothetical scenario, but it's that's built on real documented attack techniques. Right? So prop injection, specifically, is the number one exploited vulnerability in AI systems per the OWASP twenty twenty five. Right? Indirect injection, which is, like, where there's hidden instructions and documents, has been observed against platforms like Microsoft Copilot, GitHub Copilot in production. And in a scenario that we'll walk through here, Tom, the agent doesn't, like, hack in the traditional sense. It doesn't need to. Right? It starts inside a normal business workflow. So let's let's work through this. Right? There's four steps that I wanna call out. Right? The first step is poisoned input. Here, the the agent reads a a vendor ticket, an invoice, a web page, maybe a campaign brief. Again, this is not a a rogue agent. This is one that the business set up to carry out some type of business critical task. Okay? In this case here, though, this support request, it contains some type of hidden instructions. This is the poisoned input. Right? And it's invisible in most cases to any human viewer. Alright? So that's step one. Step two is what I call excessive agency. Again, the agent is doing exactly what we designed it to do. It browses. It retrieves credentials. It completes admin tasks. That's the excessive agency. Right? The agent has more capability than the task requires, and there's no guardrails on the scope. This is why for years time, we've talked about in security how important it is to to use least privilege. Right? Like, what does it need to do, and let's give it the minimum amount of privileges it needs. Alright. Let's keep moving. Step three. Step three is then the agent reaches a disconnected app. K? Maybe it uses a stored credential, maybe it uses a shared password. Either way, the action is happening outside the identity control plane. And where does this all culminate? Well, let's look at step four. Step four is this is what I call the the quiet state change. So maybe the agent creates a backup admin, maybe it changes a payout destination, maybe it adds a webhook, it exports data. Right? From your IDPs perspective, everything might look fine. But what it doesn't see is that downstream action inside the disconnected app. And this is exactly where AgenTeq AI turns an identity blind spot into executable workflows. So, again, the gap was already there. It's just it's now an agent that runs at machine speed. And what's level set here? This is not FUD. This is not designed to scare people. This is a realistic scenario. That's right. This is this technique that I just walked through here, this is a documented attack technique observed in production. Right? Per prompt injection is very real. Indirect injection, it is real. Bottom line is that agents with access to credentials and disconnected app, that's the majority of enterprise deployments right now. Right? Just further data here, Tom. Right? Panaman also found that seventy seven percent of organizations had a security incident tied to disconnected apps in the past two years. And fifty five percent, over half, rely on password only authentication for those apps. So this specific scenario exploits this in the average enterprise. So the breach didn't start with AI. It ended in a disconnected app that nobody governed. Yeah. And I I love this image too. Like, you can just you see this guy's face here, and I think we've as security practitioners, we've all been in this spot here. You know, we hear about an incident, and then we we start to dig in to find out, like, what's the root cause, and that's exactly it. Right? The AI was the vector, and the disconnected app was the vulnerability. And and the hard truth, Tom, is that most organizations, they can't answer the question, like, what disconnect what disconnected apps do my agents have access to? Because they don't know what those apps are, and that's exactly where govern come governance comes in. Great time to have another audience participation moment. Here is a second poll, and let's make it personal, Matt. Question for the attendees. How many of you have an agent right now that touches an app that is not governed by your IDP, answer honestly. Yeah. That is such a key point here. I love always love to remind people, if you're just gonna answer the way you hope it is, like, that doesn't that doesn't help. Like, just be honest. Right? Like, like, what is it right now? Do you know? Right? So, again, our our answers here, our server results, they're gonna come back, you know, one of two ways. Either it's gonna be a low number or people are just like, I have no I have no idea. That uncertainty, that is the problem. If you don't know, then you have ungoverned access that you can't account for. Now if the number's high, that's also the gap. Right? Like, that is the gap right there. So what I wanna point out from these results here that we're looking at, right, this is exactly why you govern before you automate. It's not just a framework. Right? It's a starting point for a conversation you need to have with your team this very week. What do you tell a security leader who comes to you and says, I'm already behind. My teams are deploying agents. What do I do? And we know that's happening whether they're seeing it or not. Yeah. It's great. And and I literally get that question probably three to four times a day every day. What I like to say is, like, don't automate the blind spot. Right? And there's a lot there's a lot under that. We'll walk through this and, you know, I like to break things down three to five steps. Right? It doesn't mean that agents should never touch disconnected apps. Right? That I think that would be very unrealistic. Disconnected apps, they're they're growing. They're not going away. It's not a melting ice cube problem. Many important business workflows, they already depend on apps that sit outside your IDP. The point is is that you gotta govern before you automate the work. So before an agent touches any business apps, there are really three key questions that I believe questions are more important than answers. So this is where I would start. Right? First question to ask yourself is is can my existing identity stack, right, the existing set six, seven, eight figure investments that I've made, can they see the apps? Right? Is is access tied to a known employee role or workload identity? That's the first question. The second question is, can access be managed centrally? Right? Can we provision, deprovision, rotate credentials, and force controls, and remove access without manual work. That manual work, Tom, that that's where in this world of agents, it it just doesn't work anymore. And then third, can the actions of that agent be audited? Right? So if the agent were to change a setting, if we were to create an account, export data, add a webhook, anything, can we see what's changed, who triggered it, and how to revoke it? Right? So that this is this is the real readiness test. Right? If you can't answer yes to all three, adding an agent does not make the process safer. It actually makes the blind spot faster. So govern the app, govern the access, and then automate the workflow. Okay. Sounds straightforward. We can agree to it in principle. But what does governing a disconnected app actually require? There's there's there's a lot of pieces to it, and and I think it's a it's a great question. What what I typically come back to with leaders again is just, again, three key questions that we're looking at here. A, do you know the apps that are disconnected from your IDP? Most don't. They they kinda have a loose sense in terms of what those apps are. Maybe there's two or three to come to mind. But, again, if we go back to the earlier part of the presentation, Tom, we found that on average, organizations have eighty plus of these disconnected apps. So do you know where those apps are? And then how do you get those managed by your IDP? So it's not, like, stop using agents. It's close the gap first. And, again, agents on governed apps, those are a competitive advantage. But agents on disconnected apps, those are a liability. So, Matt, walk walk me through what this looks like in practice. How are you actually helping your customers connect to their existing identity stack? Yeah. That's a great it's a great question. And the keyword there is extending. Right? You don't you don't rip and replace anything. So we purposely built Cerby to plug into what you already have. Right? And just to familiarize, our attendees with this diagram, left hand side of your screen, this is your existing identity stack. This is your your Entra, your Okta, your Ping. It's your SailPoint, your Saviant. Right? This is your existing stack on the left. You've got your agents on the top. On the right hand side of your screen, these are your connected apps. These are the ones that we know about. They work great with your IDP, the the Zooms, the Slacks. They they support all the standards. Where the challenge comes in is, again, the disconnected apps that we've been talking about. And these the the disconnect app use case is is very broad. Right? You could have something as seemingly benign as enterprise social media. Right? Think every organization, every person that's watching this webinar, your organization has social media credentials that represent your brand to the entire world. K? So that that's an example of a disconnected app. Right? Marketing teams, PR teams, those are just username and password apps that many people have access to. It's outside the governance of your existing identity stack. And then you do have legacy apps, things like that. This is what we're talking about when we say connected versus disconnected. We built Cerby to extend your existing investments. You don't have to rip and replace. So all those capabilities that your existing identity stack has, Cerby extends those capabilities seamlessly to those disconnected apps, allowing your agents to authenticate to your existing stack and then use that authentication and those grants and those privileges on those disconnected apps. So, again, it's about extending, not ripping and replacing, and Cerby plugs into exactly what you already have in your authorization layer. So now let's give them something they can do immediately tomorrow. What do the first ninety days actually look like? Yeah. So what I like to say is that this this you know, given that agent's move at such a rapid clip, this doesn't need to be a multiyear program. Right? You can make real progress in ninety or a hundred and twenty days. Depends on the size of the org. But let's just let's I wanna give you something that is concrete. Right? So let's break it down. Right? What do I think is is realistic for most organizations in the next thirty days? Well, here's what I do. Right? You gotta start by knowing what's outside your identity control plane. Find the disconnected apps, rank them by risk, start with just your top ten, and then add them to your risk register. Right? You you know, we've been saying this for years in security that you can't fix what you can't see. It's very true. Right? In most organizations, Tom, they haven't even done the inventory, or maybe they have some kind of loose idea around it. So that's that's step one. Right? And there's a lot in there. And, if you are a, you know, massive multinational with two hundred thousand employees, this might take longer than thirty days. Right? This could be ninety days right here. K? But just general guidelines. The second piece, this is where you define what I call the operating model. Start measuring all the manual hours that your team spends on managing these disconnected apps. K? This is where you've gotta define what your joiner, move, or lever process looks like for disconnected apps. What does it look like? You're gonna have to go and look at tickets. Hey. What happens when someone or an agent someone requests access for an agent to a disconnected app? Does it get logged in a service ticket? Is there emails that come in? Like, figure out what that manual process looks like. And then specifically pay attention to your biggest offboarding failure points. Right? Those are your typically your highest risk apps right now. Once you've done that, the last step is to what I call validate the model. Right? Just pick three to five apps, apply one standard control workflow, and then measure the the gaps that are closed in the time required. That's really your business case for scale. And then, you know, we would love for people to kick off their thirty day Cerby pilot. Right? That's really kind of the call to action here. So if you wanna accelerate this, we can kinda help you go from inventory to governed. And I think this is a great thirty day plan, for thirty ninety, one twenty day plan for organizations that are struggling with this. And we know, Tom, that every organization is. Yep. Absolutely. Well well said. Now we have the opportunity to take some questions from the audience. I wanna remind our attendees that if you have questions, you haven't submitted them yet via the chat window on your screen, you can. Any question we don't answer today, we'll get a response back to you via email. But before I go to questions from the audience, I have one more question for you. For the security leader watching right now who's already deployed agents, what's the one thing they could do, should do, need to do as early as tomorrow morning? I would ask your team one question. What apps do our agents currently have access to, and which of those apps are not managed in our ID pay? Simple question. What apps do our agents currently have access to, and which of those apps are not governed by our IDP? If you can't answer that question, if your team can't answer it, you have a governance gap. And that gap is your attack surface. So start there, inventory the disconnected apps or agents touch, then govern them, and Cerby can help. But even if you don't use us, just start with that question. Matt, before I get to the questions, what are we showing our audience on the screen right now? So this is a a a report that we did, very recently, and it is a, IAM buyer's guide. It just talks about how again, some of what we talked about today that most organizations are assuming because they've made investments, six, seven, eight figure investments in an identity stack, that that immediately means they've got full coverage across all their ass. But as we talked about today, that's not always the case. So what this report does is it's gonna help you kind of understand this a little bit more, and it'll give you some ways to help quantify that risk in your organization. So scan the QR code, download this report. If you have questions, as Tom said, feel free to ask them. If we can't answer them live, we'll we'll get those answered over email. Well, fact, I have some questions for you. You ready? I'm ready. Let's do it. I love questions. The best part. Alright. Number one, agentic AI sounds powerful but also risky. How do you balance autonomy with control and identity workflows? It's a good question, and we talked a little bit around this today in the in the webinar. You know, when it comes to securing your agents, when it comes to securing your, your disconnected apps, it's it's really important. You know, it's it's popular to say protect AI with AI, and I think that's more of a marketing slogan right now than it is read the reality. Right? The truth is is that when using AI, it's gotta be used surgically. It doesn't replace deterministic controls. One of things you might have heard about, Tom, right, is you hear about this concept that AI is nondeterministic. It just means, right, if you go to you you can see this very easily. Go to ChatGPT, ask it a question. It gives you an answer. Next day, ask it the same question. Give you a different answer. Right? Third day, different answer. That's nondeterministic. So it's really important that when you're applying AI to some type of security control, it's done in a nondeterministic way. So great example. Right? So I mentioned briefly what we do at Cerby how we extend your existing identity stack to disconnected apps. K? We use AI on our platform, but we use it only for basically creating these connectors. Right? We don't use it to run it in real time to carry out these identity and access management options. Right? And you wanna have a human in the loop. Right? We do that at Cerby So use AI where it makes sense, but don't use it as a way to to replace deterministic controls. Matt, second question I have for you. How does Cerby actually secure apps that don't support SSO, SCIM, APIs? Yeah. So, you know, if if I won't flip back to the slide right now, but if you run back through it, there was the diagram I showed where it's like, here's your existing identity stack, here's your disconnected apps. The best way to think about how we secure those apps is the gateway into those apps is predominantly identity. And because they're disconnected, that means, again, they are using they are their own little island. Right? They're using username and password. There's probably not to a factor authentication turned on. You don't know what the permissions that are assigned. Right? Cerby extends all the governance, all those processes that you've already built in your identity stack. We make it so that you can extend all of that to your app. So that's both on the authentication side as well as, Tom, on the authorization side. So automatically provisioning users, deprovisioning users, gathering the roles and entitlements for those disconnected apps so that you can manage those back in your existing identity stack. That's how Cerby does it. So it's not rip and replace. It's extend. Well, segues to my third question. How are identity vendors using AI in their products today to address these disconnected apps? So I think we're still, you know, just I would say this is applicable overall to AI in in cyber. Right? We're still in the early days. Right? Using a baseball analogy, I think we're maybe in the second inning. Like, it's still very early. I think the important part here, we touched on this a little bit earlier, is that you wanna make sure that when you have a vendor that is that says, hey. We're using AI. And, Tom, like you mentioned, you and I were at RSA and did not did not every vendor have AI? Like, I I found three that did not have AI. Three that said they didn't have it. Yeah. Three. It was and those are the three. Had to have been outside the venue. Yeah. It was actually in the start up vet start up part. It wasn't on the main floor. You had to go to the start up area. There was three out of seventy that didn't mention it. Here here's the point, though. When your vendors say, hey. We're using AI, ask them the question. Like, tell me more. How specifically are you using AI in your product? Because what you wanna listen for is how can they do it in a way that's deterministic. And when it comes to disconnected apps, it's even more important because what they're saying if someone's saying, hey, we're fully AI, what they're saying potentially is that there could be variability in how those workflows are carried out on your disconnected apps, and that's not a good thing. Right? When it comes to audit and governance, those IAM workflows need to be carried out in a deterministic way. It needs to be the same way every single time where you were going to have massive issues come audit. Matt, last question I have for you. When do you see organizations getting the most value out of Vagintic AI when it comes to identity management today? I think, again, still early days. Right? Maybe even we might only be in the first inning when it comes specific to AI with, AgenTik identity. I think it does a very good job at parsing logs and extracting out, like, from all the noise kind of that needle in the haystack. Right? I think we're I think we're there. Where things are still very frontier and early is what we were talking about. Right? Again, how do you take AgenTeq AI and apply it to IAM workflows? Like, the operational part of it, carrying out tasks. So to answer your question, right, just where are they getting the most value right now, I think most organizations aren't. I think the value they're getting is is just kind of taking is lowering that signal to noise ratio, and they're yet to apply it in a way that is operational to their identity workflows. And that's, again, what we're trying to do, what we've done with the Cerby platform is allow organizations to remove all of that manual work that's happening today with their disconnected apps with the the all the kind of what we call that swivel chair. Right? Taking a ticket over here and then having to go to this app and do this manual work. That's a place where in Cerby we have heavily invested so that they can get that time back and leverage AI in a way that is deterministic and helps them drive real business outcomes. Great insight. Matt, I've enjoyed our conversation. They've been terrific. I appreciate everything you shared. Thank you so much. Yeah, Tom. This has been great. Thank you. Also wanna thank our attendees. We know you took time out of your day to be here at this session. We're grateful for that, and I trust you're walking away with some excellent new insights you can put to work. As always, I look forward to seeing you at one of our upcoming events. And until then, for ISMG, I'm Tom Field. Thank you so much for giving us your time and your attention today.