Webinar recordings
Every Agent is an Identity.
5 tips for governing your AI agents, from the webinar by PlanB. and Elimity.
AI agents hold access, touch data and act on their own, yet they don't show up in your HR system. When the person who built an agent leaves, the agent keeps its access.
Isabelle Pfahler (Solution Expert, PlanB.) and Christian Bartels (CRO Elimity, former CISO) show how to see, own and govern your AI agents:
1. Find every agent, including the ones built outside IT
2. Give each agent an accountable owner
3. Check what each agent can access
4. Handle owner changes and departures
5. Prioritise by risk and keep audit-ready evidence
For CISOs, IAM and security teams who need answers for NIS2, ISO 27001 and DORA auditors: which agents exist, who answers for each one, and what can they reach?
More about Elimity for AI Agents: https://elimity.com/resources/use-cases/ai-agent-governance
View transcript
And I think we're live. Hi, everyone, and welcome to today's session. We'll give it another minute or two so people still have the opportunity to join. Make yourself comfortable to get a tea or coffee as it's rather cold outside. And let's wait for some more attendees to be part of the session. My name is chiel. I'm part of the limited team. But more important, I'll be the moderator of today. But I'm very happy to be joined today by Isabelle. Isabelle is part of our Limited's newest partner in the DACH region, Plan B. And on the other side of the screen, you probably see Christian, Christian Bartels. He used to be a CISO himself as well. But as you know, he's now working with security teams to make sure they have complete visibility regarding their identities. Maybe first question to you both. How are you guys? How are things in the exotic Frankfurt, Germany? I think, in Ghent, in Belgium, Christian? Christian Bartelsen- Hey, Braun, actually. Good. We've actually seen a little bit of sun today. Highly unusual for the German wintertime, I feel like. So far so good. It has been a long day today. So excited to close it on a high note with the webinar. So happy to be here. Perfect. And you, Christian? Christian Barte- Indeed. Fantastic. Now, looking forward for the session. I think we have some fantastic topics coming up. Great use cases. And yeah, weather. And it's Belgium winter weather, right? Dark, cold, and wet. Exactly. Talking about winter. We'll do a bit of housekeeping. But in between, feel free to drop in the chat where you're based and what the weather is these days. And then we'll go over a bit of housekeeping. So first of all, everybody who registered for the webinar will receive the reporting automatically after the session. And also the slides we can share with you on request. If you have a question, usually there is like a chat somewhere where you can drop the questions. And at the end, thank you, Christian, rainy Belgium. At the end of the presentation, there will be a dedicated slot to answer all of your questions. So if you have a question, drop them in the chat. We will take a note and we will answer them at the end. And the webinar will take maximum one hour. If you have additional meetings, there is no issue at all. So let me take you towards, Hapike, to the agenda of today. So as you know, the topic of today's webinar is every agent is an identity. And we all know, we all realize that these AI agents are arriving quite fast in organizations. But we see that there's a lot of security and identity teams that are still struggling to know which agents has access to what's within the organization. So that's actually the topic of today. And so we'll start with a quick introduction of both organizations, both Limiti and Plan B, of course. And then Isabelle will take us a bit deeper into the knowledge. We will tackle some challenges they are seeing regarding AI agents, but also some actual crystal clear use cases in the European market. We will be talking about actual customers, which is the most valuable and most concrete, I think, for the audience. And then afterwards, Christian will take it over and he will highlight a bit what Limiti is, what the solution can do. But we will not make it too commercial and we will not go into details about the product itself. As mentioned, if you have a question, drop it in the chat and then afterwards there will be a Q&A. That being said, maybe short notes on Limiti. So Limiti is a Belgian company. We're based between Antwerp and Brussels and we're based in Mechelen, of all places. And we are actually recognized by Gartner as one of the leading vendors in the identity, visibility and intelligence category. So the IVIP. And we work with security teams, identity teams, identity and access management teams in the Benelux market, of course, at the home turf. But also we have customers in the Nordics and a lot of customers nowadays as well in the DACH region. That being said, short and sweet introduction. Isabel, are you ready for Plan B's presentation? Of course. Thanks so far. Yeah, so quick word on who we are. So you know where I'm coming from. Plan B is a German technology and consulting company. And we call ourselves an AI company and a trusted advisor for enterprise AI. Currently, we are, I think, over 180 experts across AI, modern work, security, cloud platform and software engineering. Constantly growing, which is very nice to see. And our job is to help organizations basically turn technological innovation into solutions. Our principle is more, we start with architecture and processes and requirements. And then we find or we build the right tool that fits the gap identified at the customer. And that works for AI, obviously. We see lots of challenges in the field of AI, which is why we started to work with customers to build the foundation that lets them operate and control and also audit AI agents at a large scale. Which brings me a little bit to today's topic, because I think AI agents have moved into the identity space quite quickly. And identity has been a core principle of security for a long time. And now there's new challenges arising. And that's kind of what I want to talk about today a little bit. The challenges we currently see at our customers already and also a couple of use cases to make it feel a bit more realistic and to see maybe where you can benefit from the information we're providing you today. Before we jump straight back into the AI topic, though, I want to take a little step back and talk about ITA maturity as a whole, because the key message of the slide is, I think, quite simple. AI agents, like I just mentioned, are entering an identity landscape that is already a little bit unfinished. Let's put it like that. So for years, identity governance and administration has focused primarily on natural identities. So we're talking employees, contractors, partners, and we kind of know what good governance should look like. So we want to know who has access to what and for how long and why exactly. In practice, what we see is that most organizations are still working towards that goal. And I think everybody who works in that space, like the IGA space, knows that this is a journey that will never be finished. So there's a lot still to be done for the natural identities. So we are lacking application coverage. We know that connecting new systems can be slow and expensive. We know that access reviews aren't right for every use case. And oftentimes the coverage there is also still lacking. And we cannot really keep up with the ever-changing environments and everything is happening faster and faster. Then enter technical identities that we also have known for years, need to be governed. But they already expose a couple of limitations within the processes that we have set for the natural identities, right? So we have service accounts and application identities that often have lots of privileges, but unclear ownership. Maybe the life cycle isn't exactly clear, and it most definitely doesn't fit with the life cycle we've established for natural identities. And now we enter a third type, the agents. And they're even faster. And they're not just simply another user account because they can access information. They can interact with applications and potentially even execute actions autonomously. So now we have three identity categories that need governance. And for all three, kind of the fundamental questions remain the same, right? So who owns the identity? What can it access? Why does it have that access? How long should it have that access? And can we actually prove who did what with what access afterwards? Because as we all know, the regulations aren't waiting for anyone. And we are on this journey together, and it's going to be a long journey. And I'm not saying that what we've built so far in IGA is bad. Not at all. I'm just saying it isn't finished. So while IGA continues to mature, we have to make sure we include the agents. And of course, like everybody else, I've seen the numbers go around, and I put a couple in the slides as well. So I think we've all heard about the number of agents per natural identity at a company and all these exponential numbers going around online. Like I said, I put a couple on the slides, and I think one to highlight is that 82% of organizations found agents that they didn't know existed. So let's think about that for a moment. That means, first of all, these agents have not been governed badly before. They haven't been governed at all. I think it's a good thing that organizations are starting to find these agents. So I think that's also a positive note, but that also makes me ask the questions, how many more have we not found yet? And now that we know the agents exist, we move to the other numbers on the slide, and we see that agents are already part of security incidents, intended or unintended, because they can access data and applications that maybe they shouldn't have had access to in the first place. So the question isn't really, when do we start? How do we start? Do we wait for IGA maturity? No, the question is, how can we integrate the agents, the identities, the AI agents into our existing IGA programs? And they are bringing a couple new challenges to the board, which means existing principles don't necessarily apply, and we should start as small as possible. And I think the first step should always be, we create inventory, we create visibility, to then find out who should be owner of these identities. So we create inventory, we create ownership, and then we create review. So we want to see how can we review the agents, how can we review their access, and how can we have provable control over what they're actually doing. Why is this hard in practice, though? So again, and you will find this throughout the whole talk I'm going to do today, we have three principles here on the slide. And I want to have a quick look at the life of an agent. Because let's imagine we have a marketing or finance team, and they have a great idea to increase productivity on Monday. By Wednesday, they have used existing technology to create an agent and some type of local platform. They've used that agent throughout the project, they haven't had to ask IT or security, which I think is a good thing, because you want to have AI as a tool to increase productivity and not have another clunky process to onboard an AI agent. So they haven't asked anyone, nobody knows about it, they do their project, the project ends, and nobody really cares about the agent anymore. The agent persists, though. So if it's not decommissioned, it's still there, it still has access, but we don't know it exists. And that poses a security risk, of course. Which takes me back to the three points on the slide. We need to create visibility, we need to see where we have agents, and we need to assign ownership to them. All of these things are very necessary to then figure out a proper life cycle for those agents, which leads me to the control plane. So we need to have provable control over what these agents have been doing, and in which context, right? So again, we go back to the topic of visibility all the time. Which leads me to a couple of cases that I think are a good representation of what we see across the board at customers at the moment. So it's two separate use cases that I want to jump in. And the first one is in regards to a European manufacturer. So let me set the scene a little bit. It's a food manufacturer, which if you are in the same industry, you will know has to follow certain regulations. So as a European manufacturer, they have about 1,500 employees, I believe. And they are currently facing a couple of challenges. So this all started with the wish to migrate to the cloud. So they want to advance from a technology perspective. They want to be innovative. They want to increase productivity. And they want a good return of investment of that project, obviously. But the current setup doesn't really allow them to do that. Because they have no IAM or IJA tool set at the moment. So they don't have a centralized overview. If we talk about audit, et cetera, maybe they know who has access. But IT will need time to find all of that information in separate systems. Because the user administration is separated and not centralized. For the migration itself, there's a lot of preparation that goes into it, obviously. Especially because they have a quite extensive file server landscape. And these permissions have been growing for years. And I think a lot of people can relate to that. They have not been cleaned up at any point in time properly. And we don't really want to move that to a new platform now. Because that's a security risk that we don't want to have. And when we throw in AI in the future, we run the risk of oversharing quite quickly. And at this point, there's a question to ask. Do we need an IJA tool? Do we need a big project? And yes, maybe we do at some point. But we come back to the first point I made before. Before we start any of that, we should have visibility. And that's exactly what that customer was asking for. Because the customer understood, okay, forward in a productive, efficient way. How can we actually create that visibility? And that's where a tool like Elimity's identity visibility and intelligence platform comes in. Because with Elimity, we can have a very quick overview of what is actually the current situation. We only have the read-only connectors. So we don't make any changes in production, which makes initial setup quite fast. The customer can have a good overview of actual access and permissions that are granted in systems and how this has happened in the first place. And they can use that as a basis to clean up before we are migrating file server, before we are going to the cloud. So they can have a clean setup, reduce the risk, and only take what they need in the migration. And I think that's a topic that a lot of people face. It doesn't even have to be a migration. But since, like I mentioned before, we are on this IGA journey, and the coverage isn't where we all want it to be. So visibility is something we lack, not only in the AI space, but in general. And that's where a tool like that can help. The second case is a bit more AI forward, let's say. And that's a customer that wants to switch on Copilot, essentially. It's a European defense company, which, again, is an industry that's highly regulated and deals with highly sensitive data. So the topic of oversharing comes up quite fast. Because imagine you would roll out Copilot, and in the first week, an employee asks a harmless question to Copilot. And because you haven't taken care of the permissions and data labeling you should have maybe done before the rollout, Copilot will just provide information the employee should have never had access to. The employee didn't know it had access to that information, but Copilot will just provide it. And that could be salary information. That could be highly sensitive personal information. That could be decisions from boards, whatever you can imagine. And the more sensitive the data, the less you want that to happen, obviously. And my point is not that oversharing is in some way caused by AI. No, but it is made visible by AI. And if we are at a company like that, that is a very important thing to deal with before we enable Copilot, for example. And again, that's where Limit comes in. So we can have an overview of the current permission landscape and clean that up so we can prevent oversharing and too broad access when Copilot is actually generally available for all the employees. And I think those cases we face a lot today. And I think the visibility is the biggest thing we lack. And I know I've talked about the three points, but I come back to we need to create the visibility so we can then move forward to ownership and control. So that's the first step. And with that, I'm happy to hand over to Christian to give you a bit more insight into how Illimiti actually does that. Fantastic, Isabelle . Thank you so much for the insights and also welcome everyone from my side. I think it was great use cases and we're going to tackle along on that and go a bit into those details, right? Because I think the beauty of the topic and I'm so happy that we have so many people signing up and so many people calling in today is, like you said, Isabel, just about it's actually not a new problem that we are tackling over here. Actually, it's quite of an old one. However, the kind of technical depth or configuration depth that we built up over the last years is just coming up, right? Because we have those, there are two kind of agents that we talk about. The one that are more those kind of standalone agents, which are more or less non-human identities. So they are built, they have a specific purpose, they have a certain environment to operate in. A lot of people, a lot of organizations are still mostly afraid of those and, of course, they are a risk and, of course, they have to have owners and, of course, you have to deal with them. But actually, what we see, especially the last, let's say, three months or something, is the part of the, we call them the on behalf of agents, right? Everything is getting connected to everything with the MCP protocol. You connect everything to your co-pilot or co-pilot if you're a Microsoft-heavy organization is already connected to so many things, right? So the kind of agent, and that's exactly my favorite example, what Isabel said, is getting access on your behalf. Hence, if Thomas Müller has too much access as an employee, his agent will have that access. And the interesting part on that is, and that was featured in quite some discussions lately, is an agent never has a moral compass, right? So it will not judge, it will not think of our, should I have access to certain kind of things or should I do it? You give it the access to a certain environment and if that's overshared, we are acting in that kind of environment. Hence, it will give you answers. There are many of examples of customers that approached us, that they had to stop their MCP projects, their co-pilot projects, those kind of things, just because of this kind of under-the-surface oversharing. Because 99%, if we are honest, of employees, what do they do? They do their job, right? They do their daily operations, they work in the documents they need to do. There is, of course, there are always bad examples, but usually no one is clicking around and trying, when I click on that folder, where can I also go and is there some nesting going on? Yeah, an agent in machine speed will just operate in that kind of accident, like I mentioned, no moral compass, hence it doesn't know what's right or wrong. So when I ask, what is my boss earning or why this colleague is sick for so long and due to nested groups, I have access to the salary information or to the sick leave documents, well, this agent will give me a reply. And those are only internal data breaches, of course, a quite heavy one from a personal level. But think about, you give access to consultants with a consultant profile, a consultant role, where there are some nested groups and there is some legal information shared, especially when you're a stock traded company and those kind of things, right? That can have a heavily impact on your organization. So how do we solve actually the kind of old problem? Because the problem, like Isabel said, is not new, it's more or less symptoms and that's exactly where the kind of part from identity, visibility and intelligence come into play. So maybe the disappointing part for you today, there is not such a huge difference from that to what's a human identity or an AI identity or a non-human identity. Hence also the title of the webinar. So it always starts, like Isabel said, we have to see before we can act. So how to see? Of course, we have to connect to all those operational kind of sources to be able to make sense of the data, to understand where is Overshame from employees, where are agents already active, where are they to, what are they connected, what group permission roles do they all have so that we have the kind of 360 view. So we're able to see and we're able to analyze where the risks are, no matter what kind of technology you use, so that we can apply the least privileged concept. Hence, only the right employee has access to the right stuff. Hence, their own behalf agents only get access to the right context that they need. And of course, we talked about also in those examples just five minutes ago, it's also about the part of the provable compliance, right? So there is, of course, the part you want to mitigate risk, you want to increase compliancy. But of course, if you organization, and I know from a lot of people that are here are some regular visitors coming from highly regulated organization, it's not only important that you do it, but also that you're able to show that you are in control, to show that you're compliant and that that will bring that. By the way, also Gartner anticipates, so that's not our own researcher, but Gartner anticipates that 70% of the CISOs or the organizations will use such technology in order to tackle exactly that kind of challenge by 2028, which is a quite aggressive approach, but we also see that in the market of the growth of Andrew Creston also today. Plenty of people joining that webinar. So that's the part, right? So maybe just two words that you already said in the beginning. How do we do that and how can Illimiti help over there? So like I said, the idea is to connect to all the kind of data sources and to the IAI data sources. I will spend some more time over there and we integrate them to many different ways and we're even able to connect data that might be hard to connect to other systems. So it's the strength of the platform being able to connect to either really new systems that don't offer standard integration from that or really old systems or in-house developed systems so we can analyze what's going on there that we can have the analytics, the continuous monitoring, access reviews and data remediation. Then there is a quite of my favorite part I always call it of course you can also in the case of what we are discussing today fight fire with fire, right? So there are two interesting things over here. Of course we need to control agents and we need to control AI being active in our environment but of course we can also use that technology in order to do the cleanup that's needed and to have the visibility that's needed to do that. So that's why I think some of you may be visited if not it's highly recommended actually our last latest webinar with Martin founder and responsible for the product of the Limiti of our MCP layer what we connected to it because if you want to understand how your data look like of course you can do that in a Limiti and the interface is extremely powerful and extremely easy to use but if you are a huge organization you want to at scale why not connect your AI agent also via MCP to a Limiti to be able to say okay list me my top risks from when I want to roll out co-pilot where are still people that have oversharing where exactly do I need to fix it to be able to gain that kind of speed because everything is about speed nowadays everything is about getting the right insights as fast as possible for the right people and then this comes really handy technology as well so the concept stays exactly the same whether we take about agent or human imagine that this is your application landscape we connect to everything over here pull the data in the Limiti platform really important to know over here we are European solution hence as Europeans what we find really important is the sovereign part of data so don't get confused by the cloud over here it can be in the cloud that can be hosted on your own premise or in your private cloud to identify all the risks that are going on there the solution is a visibility solution hence we focus on that having those deep insights into the agents where they are where they are traced hence we don't write back we don't do IGA stuff we don't do join a mover lever hence if things need to be changed it will automatically create tickets monitoring alerting dashboarding you want to be automatically in front when someone leaves and there are still agents around in the environment that don't have an owner or sponsor anymore no one knows what they're used to ideally you want to actually as soon as there is an announcement someone is going to leave in a certain time that this will be flagged so you can proactively have a handover and ask that kind of person what is actually this agent doing who is going to be respondent for and all those kind of things and of course many of especially in the highly regulated business you already do that for your human access but you should also not forget for your agents is also having access reviews of course for them as well and review who has access to them and if they should still be there so really important we connect to all those kind of ones where the agents are created right so there is the entry agent ID if those are there have the Amazon have the Google part have the cloud for agents so really hosting agnostic and we're adding a lot of them on a daily basis with our integration team and really fantastic to see once kind of connect again the context of the agent to the context of HR to the context of the nested groups maybe have an active directory or SharePoint this really becomes so powerful context in order to understand what's going on to make decisions and to know what is an agent doing for what is it doing and do we still need it yes or no and then again it's important and that's the beauty again of identity and excellent management we can always boil it down again to the same principles making sure that there are no orphaned agents and making sure that we know where privileged agents are and that they are protected in the right way and do they need to be privileged or can they be there what we also see agents that get more access over time so there is also the question do want that or is it better to have a separate agent for a separate task and creating those super agents over time hygiene also role hygiene data quality and really important SOD especially for agents when you let things run automatically those are really the controls that have to have applied over there so at the end get across your agent platforms linked to a specific owner a kind of inventory of all your agents and we deflect with the stuff that really matter what's the risk is there an owner who left what is it still doing where is it coming from so that's really the part that we are covering from our side and so it's the whole concept of our data graph with the entitlements at one kind of visibility so that this is specifically just show some examples over here of course when you're interested to deep dive into that more in detail it's interesting to maybe show you in a live demo but that won't be possible in the session today with that being said there are indeed the five key tips I think if I'm not correct if I'm not fully wrong Isabel were the one who shares them right still on mute yes can repeat please yes I was wondering we had nice five tips for our audience today were they for me to share or for you to share even better all then I go on with that so if we look at that so if go home today no matter what kind of technology use and before we jump into the questions it's important to have that kind of inventory right so if you're a smaller organization or if are able to do it manually have to find those agents to be sure where they are so that can build that including the ones of course that's the challenge of the ever growing shadow IT IT is becoming so easy and self used that have to have an owner obviously I think that's why I thought Isabel would tell it because there was a lot of golden nuggets she already shared in her part but have an owner have someone who is accountable I think I saw the other day a photo of a book from IBM from the 1960s saying a computer can never be accountable they will need always to be human and from liability and accountability part have to have the owner of course check the access right especially with on behalf of like I said there is no moral compass over there plan for the change so if owners leave that are prepared for that and of course like we always say work risk based there is a lot of things to do it like we would say in Dutch a big sandwich to eat so best to prioritize step by step and take it from a risk priority perspective All right with that being said I think there were quite some questions coming in in parallel so time for Q&A Gil I give it over to your moderation skills thank Christian yeah maybe about moderation skills the last slide wasn't clear that had to present it so I would fire the moderator if I was but in the end we have quite some questions and I think it's clear that there are quite some regulated organizations joining the session because most of them are in line with some regulations and auditors maybe the first question I will launch it solutions like Omara SailPoint One Identity those kind of solutions do we have to finish it first or can we start governing agents already right now what's your take on that one so I think I want to take it back to the initial point I was making about IGA maturity I think it's great that you're thinking about it establishing an IGA program thinking about tools that will support that journey but again the agents aren't waiting and don't want to limit yourself in terms of productivity so I would say can start now most definitely because the processes cannot easily translate between the identity types you need those for agents anyways they follow a bit of a different rule set and as Christian showed so nicely the visibility can create early on and that will help in the whole IGA journey with every type of identity try to govern so I think starting with visibility and starting with agent governance is actually not a bad place to start at all in that journey yeah cool thank you makes a lot of sense Christian what's your take on it maybe good to know we have customers that are in the IGA program they ended the IGA program but we also customers that are looking into an IGA project coming to start so curious about your take on this indeed said already said it right so we have them at every different stage it really depends a bit on your organization right so if this the thing is guys are doing a lot of IGA projects but from my experience is that an IGA project is a never ending project in a way right so it's a living project there will be always things to finish so think the idea so sorry if I disappoint some people here if it's your first IGA project it's like a project that has a clear start and end date it's not like that it's like if build a house and there's always something to renovate and also always something to do so waiting for that kind of moment with an IGA project in hyper care phase probably it will not make sense from an attention and from a resource perspective but if have a maturity still need to onboard some applications here and there but this kind of visibility will help at any kind of stage of the project the most of course in the beginning and then have nothing and need to see certain things of course have limited visibility with an IGA product but typically not on the agents and the depth that we discussed today and of course later on if want to connect and then also the IGA can be an extremely relevant data source to connect to so that would be my two cents on that yeah perfect cool thank you then there is another question it's not about the typical IGA solutions like SIAM solutions can SIAM solutions provide the necessary guardrails for AI by establishing clear relationships between people and the agents acting on their behalf Isabel maybe first let's let's go to because have also a bit of expertise in the SIAM part of things what's your take on that one I think this is a very layered question so it's sadly a typical it depends answer right I think we the types of agents we're talking about because in the SIAM field I log into a travel portal and I have an agent I'm like can book me travel accommodations for up to 250 euros and I create this small bubble that the agent can work in so I have a clear relationship between myself and the agent gets delegated access for this one specific task hopefully that it then can do and then the access can be taken away so this is like a very specific use case that I think CM is great for and I think it's also still adapting to these new use cases but I think for that one yes but then also still like from a corporate side of things have to think about how deal with internal agents and with longer lived agents and these ones I think need to govern differently need to have their own identity that can be governed because they're way longer lived and they follow a bit of a different rule set so like I said I think it depends it's a very specific use case and I think that there's been great progress on the technology side of things yeah okay cool answer maybe Christian instead of going to we have also additional questions maybe that's more a bit in line with with the platform it's a question do also run access reviews on your AI agents yeah I mean it's the beauty for us with that kind of part it's if see a limited as a data platform an agent is just another type of data right so whether run an access review on please review all the people in your team and the access they're having or say okay please review the agents that are responsible for it just trickles down the line so it's exactly the same in that kind who is reviewing agents no one knows who's responsible for right so it's a bit easier let's say from a human perspective because then you can just look up the HR data record and okay of course saying that the HR data record is from high enough quality which is also not always the case but there have a certain reference that don't have with agents so sometimes there are approaches here and to guess or kill the agent and see who is crying in the organization that really depends also a bit on the internal risk appetite of an organization right so we have the ones that are really sharp and saying we don't know who it is we kill it and someone will complain and then we will know who it is and some are a bit more careful let's say with the stuff that they do yeah makes sense cool and maybe the last question because we're running not out of time but the Q&A is taking quite some time already there is a lot of highly regulated organizations in the call in the session and this one is relevant I think for a financial institution because the question is what will our DORA auditors ask us for about AI agents and what evidence can or should we show them today so if DORA auditors are knocking on your door will this be included in the scope Isabel maybe a question for I will try to the best of my knowledge we have people internally that are way more focused on the compliance side of things so if have specific questions be free to reach out and we can connect to the right people but generally I'm quite sure that should be able to provide some type of inventory on the agents you're running some type of concept of ownership of business purpose the environments they're running in the criticality like what do they have access to obviously so we come back to the visibility side of things they basically want to know do know what have deployed in your environment and can tell us that actually your agents are also acting on the principle of least privilege can tell us that there is some type of life cycle attached like are actually deprovisioning these agents at some point and generally I think in these audits nothing has to be automated from the start but need to be able to show that know where this information is at the moment and need to show a clear strategy on actually mitigating all these risks happening with AI agents perfect that's a clear answer to me maybe a very very last short question for Christian Christian we saw a bit of the platform but what kind of deployments does Elimity support cloud or on both so it's a containerized product so no matter what use or want to use the data is separated can have it hosted in the European cloud based out of Belgium for the moment or can have it in your own private cloud or in your own data center so as with no difference from a feature perspective so when have a lot of on premise sources it might make sense to put the solution next to it or of course if you're a highly regulated organization like a defense organization or something where you're simply not allowed to put this kind of critical data in the cloud we support both versions over there the three of them I would call it so on prem self hosted and SaaS yeah okay that being said that's the end of lovely presentation once again Isabel thank in our audience and I think we should do it way way more often because did great Christian for you thank as well as mentioned we will send automatically the recording so you can distribute it with your colleagues or watch it during the Christmas holidays or whatever like and also we will be able to share with the slides on requests but we might do a quick reach out somewhere tomorrow and ask if need something from us and we're more than happy to connect the dots both of thank so much it was a big audience today all of have a lovely evening in yeah across the entire Europe have a good one and see on the next session bye bye thank bye bye