Video: The Trust Fall: How Third-Party AI Agents & SaaS Integrations Bypass Your Defenses | Duration: 2190s | Summary: The Trust Fall: How Third-Party AI Agents & SaaS Integrations Bypass Your Defenses | Chapters: Welcome and Introduction (0s), Introducing the Panelists (202.666s), Security Tool Limitations (643.731s), Identifying Dangerous Integrations (1179.291s), Managing Integration Risks (1383.886s), Centralizing Security Data (1630.866s), Q&A Session Begins (1820.601s), Application Integration Risks (1876.856s), Conclusion and Recommendations (1985.131s)
Transcript for "The Trust Fall: How Third-Party AI Agents & SaaS Integrations Bypass Your Defenses":
everyone, and thank you for joining us today. I'm Sadia from Obsidian's product marketing team. And today, we're talking about something that keeps a lot of security teams up at night with large scale bridges like SalesLoft Drift and Gainsight happening in the past year. We're calling it the trust fall. How third party AI agents and SaaS integrations are quietly bypassing your defenses often without anyone noticing until it's far too late. Now let's start with where we are today. According to Verizon's data breach investigations report in 2025, thirty percent of all breaches now originate from third party supply chain incidents. That's double of what it was a year ago, meaning a 100% growth in a single year. When we look specifically at the breaches coming from AI models and applications, IBM's cost of data breach report tells us a consistent story where AI agents accelerate growth in numbers of third party integration resulting in SaaS supply chain breaches being the most common type of breach at 30% of all breach methods. Overall, this means that the attack surface is expanded, and most organizations aren't equipped to see it. Now let's talk about the impact that this has on your organization. When a SaaS supply chain breach occurs, organizations take an average of hundred and ninety six days just to detect it. That's over six months of an attacker operating freely inside your environment and another seventy one days to contain it. And the price tag is very expensive with an average of $4,100,000 per breach, a per third party breach. The combination of long dwell times and high cost makes this one of the most damaging threat categories facing enterprises today. The question is no longer whether you'll be targeted, but whether you're gonna see it coming. Now to understand why this is happening, let's zoom out for a second. Historically, security has always evolved in response to where trust is placed, and this has traditionally been in protecting the perimeter. We start off many years ago looking at endpoint like your mobile phones and then to network and then finally to identity. Now we're in the next chapter, which is integration security. The relationship isn't just between a user and the application anymore, but between apps and AI agents speaking to APIs and tokens. And so far, our security models have not kept up. So what it what does it take to secure this new landscape? It requires a fundamentally new playbook built on four pillars. Firstly, into inventory and monitoring. You need to you need continuous visibility into every integration or the application API and nonhuman identity in your environment. Simply put, you can't protect what you can't see. Second, control access, which is enforcing list privilege and eliminating risky or unused connections before they become entry points. Third, detect a detection compromise detecting compromises. Identify abnormal SaaS and integration behavior in real time, not six months later. And fourth, responding quickly. Understand the blast radius, contain access quickly, and remediate fast. These four things are the foundation of a modern integration security program. Protecting the SaaS and AI supply chain can't be done with just point solutions or silo tools. Overall, it requires compounding intelligence across a network where your models are able to learn from large amounts of data and empower you to map and understand not just the apps that exist, but how those apps, identities, tokens, and data interact with each other. And now I'm very excited to introduce our panelists for today, Tom Smith and Chris Fuller. Tom is a solution architect with over twelve years of experience working in identity, security, and fraud. He has worked with organizations such as Okta, Visa, and Forder, helping companies design and secure modern identity architectures. Tom is particularly interested in OAuth, OpenID Connect, and improving developer experience. Currently, he works at as a a solution architect at Obsidian Security based out of New York. Chris, on the other hand, has spent the last fifteen years in cybersecurity working in customer facing solution engineering roles at Acomi and Shape Security before moving into product. Now he leads product for Obsidian SaaS supply chain security offering focused on building solutions that solve risk coming from both integration and third party app connections across the entire SaaS stack. Now let's dive right into our discussion. Alright. So, Chris, let's start with the big picture. We talked a little bit about the trust shift. How would you say that trust has shifted over the years, and why is this conversation so urgent today? I I think I'd like to start, Sadia, by echoing the point that you made a second ago, which is that we've spent a lot of time as security professionals in implementing controls around human identities, which is exactly the right thing to do. We've ended up with, phishing resistant MFA and controls on endpoints to make sure we're controlling that human access to applications. Unfortunately, attackers have realized this, and we've seen with the recent breach announcements at the end of last year, how attackers are now leveraging these integrations and SaaS to SaaS connections in order to exfiltrate data from hundreds of different customers. So I think it's important that we recognize the fact that we now need to pivot some of these controls and make sure we're understanding these integrations. And then the second point I'd make is a lot of our effort right now as it pertains to the way we connect SaaS applications to each other is focused on point in time assessments. So often organizations that I speak to have very rigorous TPRM processes. They're assessing these third parties for their external posture, their compliance, etcetera. But once they're connected, once that token has been granted and the scopes have been assigned, they tend to very rarely, if never, look at those integrations again. So we need to make sure that we are continually assessing these integrations to your point based on their behavior, based on their scopes, net privileges, and any anomalous activity. I think the second point I'd make as well along those lines is that we've seen this increased adoption of SaaS applications, which is driving these interconnections. But also now, of course, we're seeing the emergence of AI agents. And even as I consider my workflow today, a lot of that is driven by not necessarily me logging into directly to into SaaS applications anymore, but using tools like called co work to go and pull data for me. So we're seeing this explosion not just in these SaaS to SaaS data exchange connections, but also in agents, AI agents taking autonomous action for us. And that's, again, driving the number of integrations. What it a good way to think about it is we've spent a lot of time guarding the front doors, but once we allow someone into our into our premises, into our house, really, have free rein to all the data and rooms between it. So we need to make sure that we're extending those controls within the organization. And I I said, I think this is important to remember that it's it's urgent because we've seen how attackers recognize this weak link and how the controls we're putting on these integrations maybe aren't sufficient for driving continuous security. And one thing I ask my customers, and we can ask the attendees today, is how many SaaS applications and interconnections they feel like they have because often the answer is very, very surprising. Alright. I think that's a really powerful analogy, Chris. I think it's an important point to make that adoption has definitely accelerated the number of integrations and associated risk. So my question is, like, what does it look like when one of these integrations are compromised? Like, maybe, Tom, could you take us through an attacker's playbook? What are the common gaps that they would exploit within the supply chain, and what are things that customers should look out for when they're trying to protect us as supply chains? Sure. So the the main gap in place is that really all of these supply chains are generally built on OAuth applications, which are, of course, a kind of, nonhuman identity. But in terms of understanding the gap, I think it's helpful to think about it in relation to, human, authentication. So in the bad old days of of human authentication, you know, we used to put in just a username and password, and that was enough. The the application we wouldn't have to do any MFA or anything like that. The application would just kind of accept us and the password and grant us access to the application. Now we've since tremendously improved that in the enterprise by adding things like MFA, centralized single sign on, context evaluation so that authentication for humans is much more, secure, than it used to be. But for OAuth applications, none of those controls are in place. It's kinda just like the bad old days of human authentication where the OAuth token, the the bearer token is just accepted without any further challenge or analysis. The token itself is enough to give the application access to the the API. That that's really the the essential gap is that the authentication, works fundamentally differently in OAuth than it does for human authentication. So in the SalesLoft incident, for example, an attacker was able to get access to, bearer tokens. This is kind of like just getting a bunch of, human passwords in the bad old days of human authentication. So when they presented those OAuth tokens to the API, they were just accepted without any further analysis or challenge. And, unfortunately, that's just the way that OAuth authentication works today. And in terms of the the speed of access, it's really kind of a a bad news, good news, bad news situation, because the bad news is that once the attacker has that bearer token, they can get immediate access to the API or the the application. The good news is that once they do have that access, to go go back to Chris' analogy of, you know, the rooms in the house, it's generally going to take them some time to kinda rummage around in the rooms of the house and, you know, find they might find a a a some jewelry. They might find some cash. They might find a ring of keys to an apartment complex that, one of the residents of the house owns. So that generally takes some time. But, again, going back to the bad news, unfortunately, and and picking up on Chris' mention of of AI, AI tools are making that process of reconnaissance much, much faster and much, much more automated. So, you know, on the defensive side, we've really just gotta be, kind of conscientious about making sure that we have as few of those applications, of those dormant applications around and the applications that we are allowing access that they've got access to as few rooms in our house as possible. So that that's kind of the the fundamental gap and, what's happening with some of these attacks. Cool. I think I really like the follow-up on the analogy about protecting all the gaps in your room. It sounds like there's an alarming number of ways that you can be compromised, but I think what we're seeing today is, like, a lot of organizations say that they have seams or CASBs protecting their systems. So why do you think that even after having these tools, they can't catch gaps that are, like, breaches that are going on. Sure. Sure. So in terms of SIMs and CASBs, you know, both are extremely effective tools, but they're they're challenged just because of that security gap that I that I mentioned earlier. So that the bearer token is that represents valid access to an application. So if you have that token, you have access to the application. So the API or the application, when they see that bearer token, they have no way of knowing if it's been stolen, Mhmm. if it's coming from a proper origin. They're going to accept it at face value. So they're not gonna put anything into the logs that a SIM could potentially pick up on. It takes a lot of kind of hand analysis, if you will, of the logs to see some of these patterns that emerge in these attacks, and that's why it takes so long for some of these attacks to be discovered. So, again, there's nothing unusual. There's not there there's no reason for the API or the application to question any of these tokens that are presented to them because there's no way for them to know that they are in fact stolen. So that that's part of the challenge on on the SIEM side. On the on the CASB side, they're perhaps even more challenged because often CASBs work by having, you know, an agent on the endpoint and then a proxy in front of the destination application. In the case of, you know, app to app communications or SaaS to SaaS communications, you know, neither of those elements are going to be in place. You're not usually going to have an agent on the endpoint, and you're not usually going to have a proxy in front of the API. The OAuth app is going to send that token directly to the API of the, of the service. So, you know, the CASB is really going to be completely blind to that activity. So the tools we have today are really, you know, not, sufficient, and it's part of the reason why these bearer tokens are so attractive to attackers is because they can't generally be attacked, certainly quickly by, Sims or or CASBs. Yeah. I've been I think that's a really good point to make. I think what we're seeing today is that in the past, we had a tech methods like social engineering where someone would, you know, try to call another person and try to get their credentials. But now when you can compromise, like, a bearer token, for example, like, that becomes a lot easier when it can't really be detected compared to social engineering. I think another thing that we see when we speak to, you know, our customers or people in the security space is that a lot of their third party integration security strategy is built around looking at vendor questionnaires or annual reviews. And I think we are seeing that this is not as good of a strategy anymore because this leaves a lot of gaps. So what do you think a seesaw needs to fundamentally rethink about the risk through their TPRM strategy? I think I think, Sadia, to your point, you know, TPRM process is still very critical, and and we're certainly not suggesting that customers move away from that. But as you say, they're very much a point in time assessment of the external posture of a of a vendor of an application. And once you've done that, even the the the customers that I speak to have the most robust programs are probably performing that maybe four times a year for their most critical applications, and the rest of them are maybe once a year. So as you say, there can be a huge gap in the assessment of that application and ultimately what it's doing. Whereas if we look more deeply at the supply chain and SaaS interconnections of our environment, we need to be doing that on a continuous basis. We need be understanding what scopes have been granted, whether scopes are being used versus just requested, whether the application is suddenly performing a a bunch of different activities today that it's never done in the past. So that process is very much ongoing. And, again, the way that my customers are starting to think about this is, again, performing more proactive, reviews of these applications based on their scopes and permissions, and then also monitoring them for anomalous behavior. Did this application suddenly start executing readwrite permissions, whereas in the past, it only ever used read permissions? I think the the good news here is, you know, based on on the news, unfortunately, we've seen recently that leadership is starting to realize that this needs to be a more continual proactive program and not just a once in time external posture evaluation. So I think, really, if you're trying to frame this as a as a security team member to leadership, I'd frame it very much in terms of that being this being a business issue. We saw, you know, the number of organizations who are affected by these breaches. I think there was 600 in one go just because of the nature of these token access patterns that Tom's mentioned. So it's not necessarily about preventing the business from operating, but it's more about enabling the business to safely consume these applications, monitor them, and make sure that they are ultimately acting in a safe way. So, really, I guess, to try and sum that up more concisely, I think about this in in the ways that you mentioned in your presentation, Sadia, which is make sure you inventory all the applications that you have, their permissions, their scopes, who's using them, who the application owner is. Make sure you analyze and baseline their behavior, and then make sure you perform that ongoing monitoring to see if they deviate from that baseline. Yep. Well, that's a really good way to sum it up. I think diving a little bit deeper into what you mentioned about continuous monitoring. Let's say a security team is convinced and they wanna get started. How do they even begin? Because when we look at, like, our customer environments, we see, you know, people with thousands of hundreds of thousands even of integrations in their environment and at the scale that it's growing year over year. How do you think is the best way for them to to get started? Yeah. I I can take that one, Sadia. Really, it is as both you and Chris have mentioned, it's about, visibility as the first step. So the the first step that you wanna do is take an inventory of all the the doors or the entry ways that you have into your applications. And you'll want to, of course, prioritize your your crown jewel applications, you know, the Microsoft, Databricks, Salesforce, etcetera. Now this can admittedly be a bit of a challenge. You know, one of the challenges that we have just as an industry, really, is that all of these underlying platforms that you can build OAuth applications on top of, they all name their. their integrations slightly differently. So in Microsoft, you have to look for app registrations and enterprise or app registrations and enterprise applications. Salesforce, these are connected apps. Databricks, these are, you know, service principles tied to, OAuth authorization. So it's a little bit tricky, it, you know, can take some digging just to even find out where these apps are are stored and represented inside the underlying platforms. But, you know, maybe just pick one to get started. Pick Microsoft. And to maybe, you both mentioned this, but kinda brace yourself as you get on this journey because you're probably going to find more applications than you expected in some of these platforms. So in Microsoft, for example, it's it's not uncommon to see hundreds or even thousands of applications that have been built. And so, again, first part is let's do an inventory. And as part of that inventory, you really wanna pick up as much information as you can about each of those applications, and those are key things like what kind of privileges does the application have in the underlying platform, who installed it, so who initially authorized the installation of the application, Who in your organization currently owns the application? Who can make a decision about descoping or even decommissioning the application? When was it, originally installed, and when was it last used, and when was the token last rotated. So all of that, you want to try to pick up as part of your your inventory. Not all of that's going to be in the underlying platform, of course. You might have some of that information stored in other aspects of your ecosystem. But really just, you know, start with one application and and get a handle on it and try to get as much information as you can about all of those applications. And, again, just just brace yourself because you might find, quite a few in that underlying platform. Yeah. I mean, I think definitely inventory and knowing what you have is the first place to get started. But, you know, what what you mentioned, it can get very overwhelming when you see, like, you have thousands of integrations. So how do you, you know, cut through all that noise and identify which integrations represent genuine danger. Like, what does a dangerous integration look like versus an acceptable one? Sure. So one piece of of low hanging fruit, if you will, is to just look at the the inactive applications. Oh. You know, I'm always surprised when I look at a customer's environment, and they have, again, hundreds or even thousands of these applications. But the majority, if not the vast majority of them, are either completely dormant or have not been used in in a very long time. So that is kind of initial filter that you can use to kind of reduce the size of your problem. And, of course, you know, it's just human nature. These applications, either, you know, a developer or a business user installs one of these applications, wants to kick the tires on it, accepts the the default permissions and things. So I'm I'm just gonna try this out. And, you if. I don't like it or if I if I don't think it's worth anything, I'm gonna come back later and and remove it or decommission it. And, again, human nature, that, is just not what what most people do. So these applications just tend to accumulate over time, leaving that door open even though nobody is actually using that that application. So doing a simple filter on inactive applications is usually a a good place to start. Mhmm. The other dimension that you want to look at is what permissions does this application have in my underlying platform, and does it really need all of those permissions? So, again, human nature, many of these apps have been developed by, you know, developers and companies that wanna develop something quickly. Hey. We wanna build something against Salesforce here. Let's just ask for all of these permissions, get something up and quickly, and then later on, we'll go back and descope the application so that it doesn't have all these broad permissions. And, of course, again, human nature, they get the application up and running. It works. They start providing value to their customers. And all of a sudden that, you know, those good intentions of descoping the applications, gets pushed down the priority list. That's that's how these permissions kind of get initially created and kind of stick around. So I think that we all need to be better not only about, making sure that these applications don't have, permissions that they don't need, but also pushing back against these vendors and suppliers and saying, you know, why does why is your application asking for all these scopes when it clearly doesn't need these these scopes? And, of course, looking at those OAuth scopes and making sure that if they have highly privileged scopes, like admin level access or access to users' email or access to files, that they do truly need it to, to achieve the things that they're trying to achieve with their their application. So I think that those are the kind of the the two filters that you wanna start with on your applications as you're doing that inventory is, you know, inactivity, dormancy. Can we shut some of these down? And you can even do a Venn diagram there with highly risky ones based on on the scopes. So those are the two dimensions that you really wanna start with as you're doing your your inventory. Yeah. I think an inactive integrations is definitely a good place to start. I think from what I've seen, there's a huge proportion of integration and customer environments that come from, being inactive and just getting rid of those to begin with, and rightsizing permissions would be a way to significantly reduce your risk exposure. So now that we know the strategy that teams have where they can, you know, secure this as an AI and just get started with all of these, I think from a more, technical capability perspective, like, what do you think is important for teams to do? Like, for example, when it comes to managing, alerts or, you know, trying to know what what is risky versus not. Like, from a technical or, like, creating models perspective, like, what do you think a team should like, a a team should have? Yeah. Yeah. I'll take this one, Sadia. I think, you know, again, one of the ways of managing noise is to tie into Tom's point about performing this process of pruning your integrations. Right? Because if you have a wider set, there's potentially more noise. So I think that really is an important first step. And I teased this earlier when we talked about the number of integrations. Tom's talked about the hundreds of thousands, but I think the one of the big problems here is also the growth of the integrations. If we look at one of our our customer environments on average, we're seeing a 120% new integrations year on year in Microsoft 36524% in Salesforce. As I mentioned, as we adopt agentic ways of working, I suspect that that will that growth will it will increase even further. So, you know, I think that that is important to tie this into that whole process that Tom's walked us through of inventory and pruning. The number of customer environments that I connect to where I observe things like the original SharePoint migration tool that was used by the customer to move their data when they first moved to Microsoft March in 2021 is is is, you know, always significant. Think, really, often the the baseline of these behaviors almost follows the same UEBA behavior that we look at for in humans. Right? What does what does this application do? Where does it connect from? What networks does it use? What permissions does it execute? Generally, I'd say that they're predictable, and you can determine that based on the nature of an application. So a calendar based application is going to attempt to recalendars, for example, and it's probably going to do that on a scheduled basis. So those deviations can be high signal, but also we need to understand, you know, maybe maybe I have a calendar tool and my organization has bought a new a new version or additional module there. So it's always worth, I would say, also making sure that you have communication flows built in with the application owners. Because if you want to go and validate the functionality or the deviation from that baseline, you need be able to. process that very, very, very quickly. So, yeah, we can understand that baseline of behavior, and then from that, we can determine what anomalous looks like. So if I have a calendar application that's suddenly starting to read write data from SharePoint, that's something I probably want to go and investigate very, very quickly. Or maybe I've always seen this application accessing my environment from AWS in North America. And if I see a new connection from Columbia, again, that might be behavior. Often, we see in our environments as developers doing something, but you can also have a conversation with the vendor of that application about whether they should be doing the development in that way. So, again, you can take that deviation. You can take the expected behaviors versus those deviations, and. then also make sure that you have, you know, this unambiguous perspective of what's going on. So, yeah, again, based on that behavior, we can generally predict what these integrations are doing and then have a perspective of what anomalous looks like. And then, again, as I mentioned, it's very, very important to make sure you have a process, identify application owners. If you do spot something anomalous, you can go and ask the administrator of your calendar application whether you've just bought a new component or whether you need to do something very, very quickly like revoke that token. I think where this becomes challenging soon is the AI agents. So the AI agents act potentially a little bit less predictably because they're going to have a task that they choose a path to accomplish. But still, I think the same processes of baselining behavior, tying those agents to integrations will will be important as we consider those factors that have introduced. Nice. And then I think how would you tie all of that data together, like, from a technical perspective? Like, you would need, like, a model to sort of synthesize all of this data and be able to tell you, like, what's happening within your environment. Yeah. I I I think just to pick up on some of Chris themes that what we want to do almost as an industry is is to come up with a, you know, really a centralized graph almost of what normal looks like for all of these applications, normal in terms of the the behavior that it they are executing against the underlying APIs and platforms as well as the scopes that they are requesting and, in fact, using. So if there's an an OAuth application across, you know, 500 different customers, not just within that same customer, that is is asking for a previous scope but is never using it, then it's time to kind of, you know, get together and ask that vendor to remove that scope from the the installation. So, yeah, that's a little bit more aspirational, but, you know, I I think that, pooling our resources together is is a good way to achieve more security here. Yep. Agreed. I think definitely, like, a combination of what you manage about the graph and compounding knowledge from all of, like, the data that we have was to go very hand in hand in allowing us to, like, baseline different levels of data to see what's abnormal versus normal. I think just a last question to the two of you. Like, if there's one thing that security leaders should take away from today's discussion, what do you think that should be? Chris, do you wanna go first? Yeah. I know we're almost at time. I also see a question in the chat as well, so we'll we'll try and get through as quickly and and take on that question as well. I think, you know, fundamentally, as I mentioned, like, make sure this is a business led initiative. You'll need resourcing and tools in order to perform more regular reviews, build these baselines. Of course, Obsidian can help with that and make sure that we have the you know, we help you with the ability to understand these applications. Indeed, our knowledge graph is taking the identities of users that integrated these applications, the behaviors of these applications. So, of course, there's a there's an easy button here, but otherwise, make sure that you're, I guess, fundamentally driving that business, that initiative using the urgency of some of these breaches that have happened and the the deep impact and the number of customers are affected. And also, you know, that continuous framework that we spoke about, Sadia, in terms of entry, these applications, monitoring them, and then or excuse me, analyzing them for their behavior, then monitoring them to see if they deviate from that behavior, both for agents, nonhuman identities, these integrations, and performing continuous validation and verification on those integrations is is important. So I suspect Tom has probably got another couple of points to add to that. But before I hand over to Tom, I'd say that we have we we have. a a guide that we're about to publish, on supply chain and agent agentic AI access to these applications. I believe we're gonna share this over email at the end. So, folks and attendees, with us today, you'll see that. And if not, you can jump onto your Obsidian Security website and request it. But, Tom, I don't know if you have anything else, to to add to that. I think you've covered it. It's really about, you know, the the inventory and and the analysis of behavior and, really, the the fundamentals, you know, making sure that tokens are not expired or about to expire, Alright. and and really just being disciplined about how these applications are gaining access to the underlying platforms. Tom, Chris, this has been a fantastic discussion. Thank you both for being so insightful into the gaps that we're seeing in supply chain and sharing frameworks on how organizations can protect themselves as we see this exponential increase in integration risk. So now we're gonna open the floor up to some questions and answers. If you have any questions, please drop them into the chat. I can see that we have one which is asking on, are there any recommendations on how we can define an application? And that's, like, one of the pieces that their organization is currently struggling with. So do any of you wanna take that question? Yeah. I'll I'll take, I guess, a less technical perspective. And then, Tom, you can probably give a give a a more accurate answer as well. Think, you know, it's it's it's a relatively simple question, but actually quite hard to do because at this point, what is an application? If you take, you know, Microsoft three six five, like Tom mentioned, you have the concept of internally developed applications. You have Microsoft components. You have third party integrations that connect into that environment. And then on top of that, each of these app application environments like the Salesforce is the ServiceNow, the Workdays of the world, each of those have a different way of thinking about these applications. So so the question, you know, what is an application? It can be surprisingly complicated to to give an accurate answer. I think the way that I guide my customers into thinking about the risk is that we tend to look at those tokens or integrations that we can attribute to a third party because those tend to have the the biggest risk. So it's those applications and integrations that we know come for external vendors, the most important, most critical, where we've seen the breaches arise most recently. And I'd say second point here is also what we're starting to notice that those integrations are more prevalent in our customers' environments that they realized. The the Salesforce breaches got a lot of news at the end of last year, but also we found that those integrations were also connecting to things like Google Works pace and Microsoft three six five, and there was a temp text for trade data that way. So we also need to understand and key in maybe to our vendor review process and understand what those integrations and vendors are that we, you know, that we use and and we kind of share across different ecosystem platforms in our environment. So that's the way I tend to think about it, but, of course, not often you have to go and look at each tool and understand what they consider an application and where it came from. Is there a marketplace, etcetera? Tom, I don't know if you'd you'd give a a or add anything else to that. Yeah. I I think it's it it really depends on the the hosting platform, if you will, or the the API. So for most of those, you know, kind of key SaaS applications, Microsoft, Google, they're going to have a section in their application that says, okay. What has access? What has access to the API? So anything that is an an OAuth application is going to be listed there. For something that is, maybe homegrown or, not quite. as defined as some of those, you know, key SaaS applications. It might be a little bit trickier, but I guess from from an abstract point of view, you know, an an axe an application could be defined as anything that has access to the to the API. Usually, that's gonna be an an OAuth client in secret, But, yeah, that that would be a good starting point. Thanks. I think we saw another question come in as well, which is that, you know, if you're a midsized company and you don't have a huge security team, is this something that you can realistically tackle with, like, a leaner team, or is it only an enterprise problem? I think part of the question is, like, can you afford to ignore it? You know, there's often a challenge. I'd start by making sure that you've covered some of the basic controls. So, you know, even understanding who can authorize and integrate applications is important. Because if anyone can connect an application to your environment, then that can become problematic very, very quickly as folks start to connect the Alexas and Garmins and open floors of the world to your environment. So I guess the first thing I'd often ask or or recommend that you do is make sure that you stem the tide of new integrations and locking down who precisely can add integrations, applications without without approval. And then, again, I think it is important if you're not really doing it as part of your third party risk management processes to understand what those critical applications and vendors in your in your environment. And then, again, I would use some of the framework that Tom explained earlier about understanding scopes, permissions, levels of access to at least perform review of those most critical applications as well. So I think to the question, absolutely, this can be accomplished by large enterprises and small teams. Just make sure you have those fundamental controls and then tackle the integrations and applications and vendors that most critical environment first. Sorry, Tom. I jumped right in there. Anything anything else that you would you would add? No. I would do back your your recommendation that it's really a a business decision, and it's really you need to have the the business behind you to, you know I I think, like I said, just stop the bleeding. Make sure that only a few people are allowed to add new applications to your ecosystem to at least stop the problem from growing any larger than it is today. Yep. Definitely. I think, that kind of sums up our, topic for today. Thank you again everyone who joined us today, on this Thursday morning. If you wanna learn more, I we have dropped the link to a, like, a guide that we have created. It's titled trusted by default and based on how AI agents and SaaS integrations are inheriting access. Please check it out. If you're interested to learn more, please reach out to us for, like, a demo as well, and we would love to connect with you. So thank you once again.