Agentic Identity: The Next Frontier of Identity Security

Great, so good morning, good afternoon, good, good evening everyone based on wherever you're joining us from today. Thank you for joining our latest Cisco customer success webinar.00:10Be talking about identity and how it is the next frontier of identity security. Over the next hour, we're going to be looking at.00:20What happens to identity security when the thing that is actually asking for access isn't a human at all. We'll cover why this is arriving faster than most identity solutions are ready for and how Cisco Duo approaches it. And towards the end of the session, we'd also be showing you a quick demo of how to how all of this works in production.00:28Let's start with a quick round of introductions. My name is Shreya Singh. I'm joined by my colleagues Angu Data Yasir Irfan and Andrew Gorizez. Together we'll be walking you through the presentation. We'd have Andrew on the panel later today to help you out with questions, and this session together we wanna make sure by the time.00:51When you walk out of it, you know what agentic identity is, what are the challenges that it's posing as far as identity security is concerned, and how Duo and Cisco are bracing to handle all of those challenges for you.01:09A few housekeeping items as always before we get started. We want to make sure that we can run this session smoothly and utilize the next 60 min in the best possible way. So we've put all lines on mute. If you have any questions, there is a Q and A panel that you would see in the bottom right section of your screen, please feel free to post your questions there.01:23We have dedicated time towards the end of the session for Q and A So we would make sure all your questions are addressed. For whatever reason, if the Q and A panel is not coming up for you, you can also post your questions in the chat because at the end of the day, we just wanna make sure you feel like your questions are being heard and addressed on the group. So that's what our.01:42Q and A process would look like for today's session. Okay, switching gears and talking about our agenda. So I would be starting with why agentic identity is suddenly so important, what the market is looking like today as far as agentic identity goes. I would then be handing.02:03This over to my colleague Anku, who will talk about what makes an AI agent different from both a human user and a traditional service account, and then works you through Cisco's approach, which is NO, authorize and adapt. From there, our colleague Yasir will talk about how Duo discovers registers and.02:22Features and then the part that all of us are genuinely excited for is a demonstration. So we're gonna show you what.02:42An AI what one AI agent driving a real network with policy deciding for action what it is allowed to do. And of course, as I said towards the end, we have dedicated time for Q and A So we'd make sure that once we cover all these topics, we're also addressing all your questions and you work out with all the information and resources that you need.02:50Perfect, so before we start looking at the solution, let's learn a little bit more about the problem. Now, I'm sure all of us collectively agree that we've been hearing about AI a lot more in the last two to three years. If, if I compare this to maybe five years ago, we weren't really hearing about.03:13About AI as much if, if we compare it to now and AI has become like this buzzword that that comes up in every aspect of our life, be it work, be it our personal lives, we are relying on AI more and more for our day to day tasks. We are relying more and more on AI for research, and AI is just every.03:30And the evolution of AI has been pretty rapid as well. So we started with chatbots, which is static knowledge, which was the 1st stage of AI, where whatever you needed in terms of a query that you had to get addressed, you just type that and that you'd get the response and the response would be basically based on what.03:51That AI model is trained on and it had a certain boundary on what of what it can and cannot do. The evolution then transitioned to a real time knowledge or a tool enabled AI. Now this one specifically can call an API, it can query a system, it can fetch.04:11What it needs to answer you properly, so this is not just dependent on static knowledge anymore. This is something that's gonna get you knowledge in real time. And the 3rd one, which is the next step of the evolution of AI.04:31Is autonomo autonomous decision making, which is also the agentic error of AI. And this one specifically is not just different and kind. It's also different in degree. Each of these steps that you see.04:45Has not just added a capability. The last one specifically has also added the ability to take action in your environment without a human in the loop for every decision. Now there is a log that you'd see on our website which was posted in March 2026 that talks about something known as the agent trust.05:00Yeah, so what we did was we surveyed IT executives and security leaders.05:20And we tried to understand how they were using AI in their environment today. Turns out 85 % of them were very confident to actually experiment with AI, right? And and they were very happy to try things out and see how AI could benefit their existing setup. But if you look at this little number on.05:26On the right hand side of your screen. There were only 5 % of these.05:46Who were using AI in their production setup, and this gap between the 85 % metric and the 5 % metric is what we call the agent trust gap. You know what was really surprising is when we ask these leaders, why are you, aren't you really making using AI in your production environment?05:50Why is it that 85 % of people are experimenting with AI, but as far as it it means that they have to make any changes in their production setup and actually use AI in their production setup, they're not very confident.06:10Surprisingly, the answer that came up was not related to cost. It was not related to capability. It was actually their lack of trusting agents. So the agent trust gap basically comes down to three concerns that you see on my screen. The 1st one is access control.06:23Which is how do we manage access control for all agents at all? 2nd one is data exfiltration. How do we prevent a rogue agent walking out with our data? And the 3rd one is unpredictable autonomy. So how do we prepare for the fact that these things make their own decisions? Agents actually.06:40Sure all of us agree are super intelligent, but if I were to ask you, do they really know what's the right or the wrong thing to do? Most likely your answer is NO. And if anything was to if an agent in in this case is was to operate without NO notion of what's the right thing to do, NO NO notion of what any consequence could look like, that.07:00We would definitely have a security implication and that's what we are trying to address today. With this.07:20I will now hand it over to my colleague Anku to talk more about agents and how we are planning to make these a part of our zero trust framework. Over to you Ankud. Thank you, Shreya for for that excellent introduction and the overview. So you have perfectly laid the ground up for why this is such a critical challenge for us today.07:26Building on that, let's take a closer look at how we actually address this risk. So, let's take a moment to look at why AI agents represents such a unique challenge for our security teams.07:47If you look at the table, we are comparing three types of users, humans, machines, and the new category which is AI agents. On the left, we have humans. So they have brought access to resources, but their speed is limited.08:01Believe they bring human judgment and common sense to their accents. On the right hand side.08:21We have machines. These are your service accounts and, and the automated scripts. So they operate with a resid predictable rules and have a limited scope, but they provide the speed and scale, that keeps our business running.08:27Now look at the middle column, which is EI agents. They take the abroad scope of human and combine it with the rapid exceptional scale of emissions, but here is the critical security gaps.08:44They like the judgment of a human and the predictable, right? Rule based nature of emissions. Essentially, an Easen has the power to act broadly and at high speed, but without the breaks of human concerns.09:03Missions like predictability. So this is why we consider them the worst of both worlds from a security standpoint. So they are in new class of digital users that we must learn to secure because you simply cannot.09:21Defend at human skill when you are, when you are operational and potential threats and are moving at mission skills. So e.g., imagine and AI isn't deployed to handle customer support tickets.09:39So it has brought access to your entire CRM like human to resolve issues. So however it operates at rapid speed like machines. If it encounters a complex case it hasn't been trained for, it lets the judgment to pause and ask for help.09:55If it gets confused, it won't stop to ask for help. It might accidentally issue a full refund to thousands of customer in seconds. So it has the power to act, but it has NO idea when to stop. Next slide please.10:14So where does this leave us? The takeaway is clear. We need to secure our workforce with the same level of with the same level of rigor that we apply to our human workforce.10:33It is NO longer operational, AI agents must become a fundamental part of every organization's zero trust framework. This means applying the principle of least privileged access to these agents they should have just enough access to get their specific job done and nothing more.10:53By integrating them into your, sorry, by integrating them into your zero trust strategy, you can harness the power of AI while maintaining the security and control of your organization requires. E.g., consider a marketing AI agent.11:14Task with drafting social media post. Instead of giving it broad access to your entire corporate network, you apply zero trust, you restrict its access, so it can only reach this specific content repository its need for drafting that post.11:32So and it has NO access to your financial or other database like HR and IT. So it has just enough access to do its job and nothing more. Next slide. So to address the rise of AI agents we.11:51We need to evolve our approach to zero trust. Historically, our security models have focused on securing access.12:11We verify humans identity using passwords MFA and device posture, and once that's confirmed, we extend trust. We rely on the assumption that the human will exercise judgments if they encounter something questionable.12:19However, with reasons, providing identity is NO longer enough because they lack human judgment and common sense. So we cannot simply trust them once they have gained access. So we need to shift our focus from just securing access to verifying every action they.12:37Day. As you can see on the, on this slide, this is a fundamental shift, for humans. It's about providing identity and trusting the users. For agents, it's about verifying their behavior and controlling the agents.12:57So we must decide on a per action basis whether what the agent is attempt attempting it's actually allowed or not. E.g., for a human we verify their identity at very 1st login using MFA or device posture et cetera, right?13:14And we trust them to navigate the system.13:34For an AI isn't identity is not enough, even if we even if the agent is who it says it is, you must inspect the actions. If the agent is authorized to read data but suddenly attempts to delete a table in your database, the system must trigger.13:38An alert or block that specific action in real time regardless of the agent's verified identity. Next slide, please. So what does a robust solution for securing AI agents actually look like?13:58At a, high level, it comes down to three key moves. 1st, we must know every reason. We need full visibility into our agent workforce, understanding exactly what exists, who owns it and what is intend intended purpose.14:17Second, we need to authorize every action.14:38It is NO longer enough to gain to to grant broad access once at the start. Every single interaction with your tools, APIs and data must be checked against policy before it happens. 3rd, we must adapt to risks in real life just because an action is authorized doesn't mean.14:41And it's safe context matters and our security decisions need to adjust dynamically as the agent operates. E.g., know every agent. So you, you maintain an inventory of all active agents, e.g., is in finance board, e.g..15:01So authorize every action. So you define a policy agent finance bot can only query the the laser but cannot initiate the wire transfer. So adapt to risk in real life. So if if that's finance bot isn't suddenly attempts to query.15:20Let's say 5000 records in 1 s.15:40Right and anomaly, the system detects the high risk behavior and automatically revokes his access until a human administrator received the activity. So these three moves rely on consistent enforcement at the access boundary. So today we are going to spend most of our time.15:43On that middle piece, authorizing every action and we will going to show you exactly how that works live. Next slide please. So, where does this enforcement actually happens? So this is where Cisco zero Trust access comes into play.16:02Our platform comprising Cisco Duo secure access cloud and Cisco identity, Cisco ICE already sits at the center governing how your users and devices connect to everything they need.16:21The internet SaaS application, private applications and your critical data and tools. So what's new in that we have added a 3rd input agents now join users and things as a primary requester whether these are custom.16:38Reasons you have built?16:58Or isn't already embedded into the application you are running or 3rd party agents. They are all now managed through the same control plane. It's it's the same enforcement point you are already using just extended to cover its new class of identity, so this is the part that our.16:59The Demonstration today is going to run on.17:19So we have covered the why and the strategic work, but to see, this all comes together, I'm going to hand it over to so he is our true expert and the key person who brought today's webinar to life, so over to you Yassid.17:22Oh, thanks a lot Anku for Andrea for setting the foundation of what we are looking through. If you could just move to the next slide. It's all about, you know, what Anku was talking about, those three things in order. 1st thing is visibility and identity management and also.17:41Who knowing the agent?17:59Because you cannot go and authorize or protect what you haven't discovered. So that's what something what I call. The 1st pillar is knowing every agent sits in your organization, among be it in on your infrastructure or be it among your employees. That's what the key pillar from an agent.18:01Security perspective is to identify those agents.18:21Let's move to next slide. Here is the position, right? Most organizations are in today. Most of us know our human users. We know that devices who are accessing all kind of applications. You know your MCP service?18:24And the resources behind them, but the layer in the middle, the agents, it's empty. There is NO inventory of them, there is NO ownership, NO policy is associated with those agents and definitely there is NO visibility without blocks. That's what.18:43Lacking. And that's the layer actually taking the actions. You cannot apply least privilege to something you haven't discovered. So, let's see how we can look into that. Now, when it comes to agent discovery.19:01Once you start looking, the 1st thing you get is a complete view of all agents present in your environment. Door discovers agent from multiple sources and monitors them for posture and configuration issues and for risky behavior.19:17And it find two kinds. The one you have sanctioned, you have authorized those agents, part of your deployment on purpose and the ones which you didn't Neither authorized, these are the shadow agents.19:33Which are equivalent of shadow IT. Except this time the shadow assets or agents can make changes to your infrastructure if NO proper controls are deployed. The middle lane was empty just 30 s ago. Now it's the busiest column on the slide. That's usually.19:49The moment the conversation changes. Let's go to the next slide and agent in discovery gives you a complete list. A directory is what make.20:09It's governable. 1st, you register the agent, then you assign credentials so that each agent has its own identity, not a shared service account, not a borrowed human credentials, yes, not a borrowed human credential. It's own. Then you.20:25You map it to a named human owner, so somebody is accountable, you assign an access policy that define what it may reach and you attach activity logs, agents, credentials, owner, access logs. That's the role.20:44Now look at the bottom row. No credentials, NO owner, unscoped access. Logging is off. That's the agent you act on 1st because nobody in your organization is accountable for what it does. It all starts from.21:03An agent directory categorizing it, that's where the action will start. Let's look into the next slide, please. An agent identity has a life cycle. Agents are created, they change.21:22They should be retired. Joiner move a labour, but those are all for software. Onboard a new agent against a template, so owner task, scope, and logins are all set at creation rather than voted on afterwards. Decommission.21:37You can obsolete agents and revoke their credentials?21:57So sprawl doesn't quietly accumulate standing access and log all of it, traceable back to accountable human beings. That's what the whole agenda is to trace every agent.22:01To a human being so that they are accountable. Next slide, share. Put together that's new agentic IAM, discovery, a directory, a life cycle management.22:18An agent to human accountability. Visibility, accountability and governance at machine speed because that's the speed agents are moving at. That's knowing every agent, but knowing isn't controlling.22:35So let's talk about what happens when one of them actually tries to do something. Let's look into the next slide. This is 2nd pillar what I call is authorize every action. So we know what our agents are, who owns them, and what they are supposed to do.22:50Now the harder question, when one of those agent actually tries to do something, who decides whether it's allowed or not? Let's look deeper into that. The three constraints on every grant.23:09Just in time, access exists only while the agent is performing the task, not before or after, just enough scored by identity context to specific tools the task needs.23:25And nothing adjacent and just long enough, short lived OAuth authentication token measured in minutes so the agent never holds a standing credential. Together, those mean an agent is never overprivileged even when it's working perfect.23:41Let's go to the next slide. Let me take the concrete. I have built an AI agent that operates a small network. There are 13 tools which are capable of listing devices, read configurations, restart interfaces, reload a switch.24:03On they can rotate the credentials. Basically, three people drive these agents, only she is a helpdesk.24:24Basically it tracks the device health and get four tools read only. Own is from operation looking after the diagnostic parts and get seven tools to operate and our administrator. She has complete control of with all the 13 tools where she can.24:33Like not only read the tasks, you can also look into the diagnostics but also take an action.24:52Now one important point, the agent itself hold NO network credentials. It inherits a short lived scope authorization derived from whatever it's driving it. And the gateway checks every single tool or tool called against the scope. This is where the scoping come.24:58Into the picture, that's where the job description are assigned to each agent. Now let's look into the next slide. Here is the shape of it. The agent never touches the network directly. Every request goes through secure access MCP gateway. By the way, you can use any 3rd party.25:17MCP gateway if you want with Duo underneath providing OAuth and per tool authorization policy.25:37Authenticate the human, allow the network path, authorize the tool against the rule, log the call, these four checks, one control point on every single item an agent is trying to do. So let's look a little deeper into this.25:45Once we go into next slide, basically our help desk operator, you know, she has got all these permissions where she can look into the devices in the network.26:02She can check at the status of the switches, she can look into the interfaces, things which she is allowed to do. But when she tried to do something, you know, out of curiosity, which is not, which is above her privilege, that's where those identity security comes into picture where it stops. I'll just stop here for the moment, you know, I.26:18I talked about how these controls can be applied. Let's see this in an action. What I will do is I will ask Shreya to stop shreying for a moment. I will walk you through the demo where you know the helpdesk administrator is trying to.26:38Carry out these tasks at a machine level. So our helpdesk administrator is basically trying to use an agent. In this case, it's I'm using cloud as an agent. So what she's doing is she's just ensuring that the character or the, you know, is active.26:55Yep.27:15So she tried to connect to that connector not connector where she wants to manage the network. This is where, you know, the agent goes through an single factor authentication and dual factor authentication where they have to type in their username and password. They can use passwordless as well if you want to. So.27:16It's all about the agent is mapped to a human here, you know? Basically when they enter the username and password, you can see, it goes for MFA. You can use passwordless if you want, you can use OAuth authentication multiple kind of authentications. Once the user is authenticated.27:36Because the single sign on factories done, the agent is get the privilege to carry out the actions. Now, the agent is connected to the network, it is connected to their noc. And here is the user, the help desk basically is asking the agent to list.27:57All the devices and the core switch, the status of the core switch in that network.28:17The agents communicate at the backend using secure access gateway, with MCP gateway and then it looks into the real devices behind that MCP gateway and brings those details to the agent, which is operated by a human being.28:21It looks into the devices, it looks into the tools, what it's running, it's getting the device status here once they have gone through their primary auth, specifically when the human is identified. So it lists them the status of their network. In this case you can see the distribution switches downgraded. If you can communicate.28:38Like.28:58Any normal prompt with this, so the user decides, the helpdesk users decides just to ping one of the devices from their core switch as source they don't need to go and sit inside the core switch, right? They are, everything is done through an agent, but they're using this.28:59Tools to simplify their access.29:19So the tools are running it's not touching the network straight away, but it's going through this and you can see the user is allowed to do this access. She can ping from the core switch and can see what's happening in the network. And out of city she want to see what's.29:22The status of an interface. So when she tries to put interface status, it goes through the process of grabbing that status at a real time and it tells the interfaces up. There are NO errors, the utilization is good. Yes, that's good. So far her is doing everything, but she becomes little.29:40What do you call, to reload the switch which is not supposed to do. As a prompt, we'll ask that, you know, do you want to really want to reload the switch?30:01So it is asking for the confirmation, so she says, let's see what she wants to do. She I think she confirms yes, go ahead and reload the switch. Here is where, you know, the agent cannot do despite of the prompts, whatever they are trying to do.30:11They cannot reload the switch, you know? Just to show it's the prompt is asking, do you want to reload the switch again out of due diligence, the prompt is doing and this time the help desk user tries to reload a distribution switch.30:30So let's see what's happened. It's blocked by an admin policy. You cannot reload this. This is where we at Cisco applying authorizing each and every action of an agent. Now let's see this in action. You know what's happening? When you log into your admin panel, you can see the user.30:47Right through a single sign on?31:07And they got a dual push, they authenticated, their primary authentication was successful. When you look into their authorization part, you can see that helpdesk owner is coming from a cloud agent and they are authorized to carry out four policies or four tools, that's what they can do. Apart from.31:09But if they try to do more, their access was blocked which we have seen, and Duo is showing you know what they're capable of doing that. Now let's see how it looks in the secure access. You know, when you go to your secure access dashboard, where the MCP gateway is configured, let's go and just to my secure access instance.31:29Which has got this things. Let me monitor the events and see those MCP gateway logs. Let me select this MCP gateway. The locks are loading. Okay, this was my network traffic, but let me go back to MCP gateway traffic.31:49Here you can see that, you know, the users, what are, what is allowed everything is shown, but there is one right color which is blocked. That was, you know, when the helpdesk agent was trying to reload the device that was blocked. No matter how.32:09The agent is re prompting itself to reload, they will fail because it's not allowed, but whatever they are authorized to do, they can do. This, in a nutshell, you know, that's how the production network will look like where you know you can control your.32:24Access to your agents, you're discovering using duo. You're discovering those dynamic agents, you know, the agents does themselves doesn't have any identity, right? You are discovering that agent is coming from cloud and all those things are recorded in both the dashboards of secure access and duo.32:42This is a short demo I had. I could jump deeper, but, you know, I don't have much of time to go there, so I will stop here and I'll ask Priya to go to next slide.33:02Thank you. Yeah yeah yeah please share your screen. Let's wait until she comes but you have seen that in action, you know what happened. So let's go to next slide.33:15By the way you know the NOC operation team, share you can click on this until then. You know, this particular knock operation guy, they are allowed to use diagnostic tools. Yes, they can do, but when they want to reload the core switch, they will be blocked, you know? Again, you know, the action or the policy.33:33Just come into the picture and their authorization will never go through and Duo will blocks them.33:53Here, then let's go back to the next role where we have Aisha, she's an administrator. She has got full privilege. She can do all the actions what she's allowed to do because she's a full administrator, she can reload the switch.33:59She has the permission to that Whenever she tries to do, she will go ahead and she will do that and she can apply those controls and we are controlling these details at a secure access level. So this is you know since the.34:15Time is limited, I cannot go into demonstrating the other use cases because the authorization token which I have will lost for at least 15 to 20 min, so that will I cannot kill those session at this at this stage, so now let's go back to the next slide.34:32What we have seen is basically, you know, the whole policy on one screen is what it's showing. There were 13 tools, three roles, helpdesk operation and admin, read everyone can read, basically from the helpdesk team operation team.34:48My admin team, but it comes to diagnostic part.35:08Only the operation team could be able to perform those tasks, right? So like running the logs or, looking into the logs, all incidents and all those kind of things when a helpdesk, user or an agent try to do that, it's blocked. Administrator has got access to all those.35:12Is all these three things, but when an operator is trying to reload the switch, NO, they are blocked. They cannot do the, the rules what administrator has. So by default it's the deny rule. The tool is not granted to take an action. It's all.35:32Applied at a group level. That's, the decision NO matter how, you know, The the re prompting or rewording of the prompt is done, it's not going to work because the decision isn't is not made by a model rather, it's made by zero trust approach of the.35:53Cisco's protecting and discovering agent toki. This.36:13Is the matrix what you know, what we saw right now? Now let's go into the next slide. It's one prompt, one agent, one server. Reload core switch only was blocked. Because she was the only one who can make things or make changes.36:19And.36:38That was enforced per tool at the gateway level before the request you ever reached even to the device, right? That's being controlled directly at the gate level, a gateway, the secure access gateway or MCP gateway rather managed by secure access, that's where it is controlled, that's where zero trust action is coming into the picture.36:39It's where you know Cisco is ensuring that, you know, even the undiscovered agents, they cannot reach to the.36:59Backend resources, whatever you have, it could be your network devices, it could be your web servers, it could be anything. You know, those agents won't become distractive and that's where you know, the boundaries are set. That's what I would say. Let's go into the next slide here.37:09One last piece is authorization decides what an agent may do, but intent can drift and prompts can be manipulated. So the gateway continuously inspects the agent traffic to identify malicious prompts.37:29Rogue agents are drift away from the agent's declared task and applies safety and security guardrails to stop it reaching your infrastructure. Authorization decides what's permitted. Real time inspection catches what's actually happening and revokes when the two.37:45Diverge. Alex ok we are in the next slide. To put in a wider picture manage agent identity with duo, agent AI, enforce agent access with secure access and prospect agent themselves with AI defense.38:05With observability through Splunk and AI ready infrastructure underneath, we can definitely have a clear picture. Today we went deep into the middle one because that's the one standing between 85 %, those who are experimenting and the 5 % who have deployed agents in production.38:24So if I can walk you through few, few logs from you know from our, what do you call this from six identity intelligence engine where you know where you can really see the agents which are non humans which are recorded our end.38:45So, if we have a couple of minutes I can walk through that and then, you know, we can jump into the.39:04Q and A part. What do you like? Yeah yeah absolutely. We can do that. Sure, let me I will just share my screen in a minute. Let me log into my Cisco identity intelligence. Okay, you know, when you go when you look into the dashboard of an identity.39:11The intelligence, this is where you know the much of.39:31Picture comes in from the human identities, which we are capable of capturing them. When I look into that, you know, you can see how many newly created users are basically the agents were there, how many of them were active, how many of inactive, and all those things will show into the picture, you know.39:34Where those agents are coming from, what kind of integration, what kind of, you know, the.39:54Source of those identity which we are recording and also you know what kind of activities they're doing, you know, that it shows me that, you know, it's coming from my single sign on MCP agent way and some of them are not reviewed which I can review what's happening there. Now e.g., if I look into four newly created.40:03It shows me that, you know, they are coming from Cloud this is the idea and this is the dynamic client identity registration. What happened, it's an application, it's onboarded, what is the date it done? All those kind of data comes into the picture. If I go back to again non human identity, you can look into that, you know, how many agents you can filter them by.40:23Agentic apps, how many agents apps were there? You know, there was cloud there in my own setup which was controlled with Duo and secure access and also the copilot, all those kind of things are coming into the picture. What kind of applications were accessed? You know, it gives you those details and also like.40:43The key what I see is which you get a visibility is dynamic registrations of those clients because you know they the user is authenticating you can discover what agent they were using, what MCP gateways they were not gateway connectors they were using, you know, using an agent, all those things and how it is registered. That's kind of visible.41:02Very comes into the picture. This is what lacking in today's solution most of the solution that's where we come handy and also you can look into the MCP part, you know, what MCP servers were used, you know, how they were kind of applications and all those things. It gives you a visibility I can go deeper in that, but I will just.41:22Somebody wants me to increase their font size, just give me a second, please. Sorry for that.41:42Is it visible? Okay. Okay ok let me just walk you through that, right? It it it captures the mcb's pro protocol what I used in my life.41:53There was one thing part and I was talking about the dynamic registrations of those clients, you know, which were used to access the applications, which tried to abuse the system also by reloading the switch, the help by helpdesk administrator tried to do that, which was captured and also the agent to caps, you know, what agent to caps were there so all those kind of.42:08Visibility comes into the picture. Let me go back to the dashboard also where you can see non human identity. Non human identity means it could be your.42:28It could be your service accounts which Anku was talking about from a machine perspective, you know, all those things lack visibility today. That's where, you know, Cisco identity intelligence is capable of, correlating the information what we have collected and then, you know, it's presented to you and then based on.42:38On that you can take an action, you can block an action, basically, that's, this is pretty much what I wanted to present right now.42:58And let me hand back for the next sure there is a question right that you know. Oh, you can use copilot. You can use copilot. You can use any kind of any any any agent.43:08Which is capable of supporting you know MCP connectors basically. So any agent, it could be Germany, it could be cloud, it could be Codex, anything you can use. You guys are using Meraki as SaaS will connect to work with that.43:26Meraki for a network perspective or if you're using see as long as it's capable of supporting If you're using duo, definitely you can use the connector will work.43:46Which I want to check, you know, what aspect of you are looking into, but I have I remember out of the blue that we are using secure connect, I feel like yes, it's capable of supporting that but further we can come back to you on that part from accounts or for your question.44:01Thank you, yasir, for that detailed demo and thank you Anku as well for sharing all those valuable insights with us. With this, this brings us towards the end of the webinar where we're taking some time for Q and A Thank you for posting all your questions so far. If you have any additional questions, now is your chance, please feel free to post those in the chat.44:18Let's see, we have a question yasir. So the agent you used to show network commands, my query was, will that agent interact with Meraki rather than duo? So that's the follow up question.44:39See it's all what I have done is I have built it this one at the backend with you, it's all the capability of your MCP server, right? If your MCP server is capable of communicating with with your with your Meraki network gear or any 3rd party devices, yes, we can definitely use.44:52For automation perspective or from a monitoring perspective or from running your NOC perspective, it's possible to do. It all depends on the capability of your MCP server what you have. Well it's all we build this lab on our own using our own resources, but you can build.45:10Your MCP server, and then it can communicate with the resources behind that. Yes, it's possible to do. I don't see any reason why.45:28It's not possible. Thank you, yasir. While we wait for additional questions, we have one more. Let's see. What Cisco license do we have for this?45:38Basically, I think if you have it's better you reach out to your account team, they are in a better position to talk you over from a licensing perspective. You know what kind of licenses you want. But definitely you need to have an access to dual Duo and secure access.45:55If you don't have secure access, you can use any 3rd party MCP gateways for this, but certainly duo is needed. But your account team can definitely answer you what tier of licenses needed to run this solution in your own environment. It's not just limited to monitoring your networking devices.46:13To be used for any kind of applications or any kind of services you are hosting on your data center or on your cloud, our SaaS solutions, you can use this is just a logic what we showed what we what for for for a simplification perspective, but it can be widely used.46:33Yeah. In the meantime, as we get more questions, I also quickly wanted to share that all these resources that we have, be it our webinar recordings, any articles, Cisco community is your one stop shop solution for all the knowledge sharing that that.46:51Takes place related to Cisco events, topics and webinars. Also, if you want to discuss this, I know we've gotten a few requests already for a further discussion with your customer success specialist, so thank you for that. But if you want to have a further discussion about this, please feel free to reach out to your customer success point of contact.47:11Request for a complimentary CSS engagement and take this discussion forward. And lastly, when we end this webinar, you would actually get a survey. This is the 1st time we've adopted this interactive way of presenting and we'd love to hear your feedback. We'd love to hear your feedback around the topic, anything.47:31That you want to share with us, please it'll only take a minute if you could fill out that survey that would really help us. And also we we do these sessions pretty often. So the next one that's coming up is in September, so it's on 17 September where we talk about breaking Active Directory attack parts with Cisco duo. So if you are.47:51Interested in joining that one as well. You'd be hearing more about that session via marketing mailers. There's also a registration posting link that's published in the Cisco community forum. We'd love to host you in that session as well. So I'm keeping this open for a few more minutes in case we have any last minute questions.48:11Okay, it looks like we're all good to go. Well, in that case, we're good to wrap up. Oh, great, we have a question. Let's see.48:30How does Agentic IM work across AWS, Azure GCP, and on premise environments? Is there a common identity model or integrations managed separately? See say that's the reason why, right? We have something called Cisco Identity intelligence, which is a vendor new.48:48It's capable of managing or looking into your different identities. Like in this case if you're having Azure intra, Google and all those kind of even for on prem environments, you know, CI is capable of giving you those kind of visibilities. I'm talking to you purely from a visib.49:09Ability perspective, not from a protection perspective from protection perspective, if all these solutions are capable of supporting you know, MCP gateways, they are capable of supporting OAuth authentication where, you know, we can rely on those tokens to control the authorization timeout when.49:28View for those agents, we are pretty much capable of integrating this with one platform or or I would not say one platform but a unified platform from Cisco which work across different different products basically from different vendors. The Cisco identity.49:48Intelligence is one thing which definitely, you know.50:08Gives you a visibility from not only from a human aspect but also from non human aspect where you just simply integrate with your IDPs and then get those data content data where you have you can take control of what's happening over there.50:11Thank you, any other questions team before we wrap up? Also, does Duo distinguish a legitimate agent from a compromised agent using the same valid identity? Oh, there are multiple ways to do that, you know? Yes, we do distinguish, there's the reason why.50:31Why, right? We are mapping an agent with a human.50:51You know, if there is a illegitimated or an agent which is not an authorized agent comes into picture where we have those specific controls where you can ensure that, you know, agent is coming from your trusted endpoint. If an agent is coming from non trusted endpoint, the access, the agent, even it cannot do its.50:54Memory authentication, it's blocked there itself. And also if you apply country specific policies, you can or authorize network policies where you can control from an authorized network. You can also ensure that, you know, not only risk based authentication, when you see your users are using the agent from different.51:14Location. Suddenly they are right now, right now if they're working from Album but suddenly you see that you know using some 3rd party VPNs or power networks, they have traveled to different places. You can block those access or some let's say that an account takeover activity happens to that particular machine.51:34You know, it's controlled by agents at the backend and who are trying to manipulate your agent by ingesting their own MCP character. Yes, you can control all those things because at the end of the day, the human either has to use accept a push.51:52On their phone or to use any kind of if they are using password like this, they have to accept or touch their fa use their face ID or touch ID or whatever.52:12To authenticate those users Yes, it's possible to control and to stop the bad actors being abusing your system. Thank you, Yasin and thank you Ron for your valuable questions. Just keeping this open for.52:23Two more minutes to see if we have any follow up questions.52:41Okay, if not, this brings us to the end of today's session. So once again, we really value all your attendees taking out the time to join us for today's session. Once again, your feedback is super valuable, so once you get a survey after this session ends, please take a minute or two to share that feedback with us.53:00Also, a big thank you and a huge shout out to all our presenters and panelists today. Everyone has worked super hard to prepare this set for this session for you. Thank you Yasir for the detailed demonstration. Thank Youngku for helping us understand more about agents and how it becomes a part of the zero trust framework.53:20Thank you, Andrew for helping all our attendees in the chat forum addressing all their queries for them on the go. So, it's all the teamwork that we did together to make this session possible. We really hope this was worth your while, and we look forward to hosting you in our next webinar on 17 September where we'll be talking about Active Directory.53:40Thank you. With this, we reached the end of today's session. Once again, thank you so much for joining us and please stay in touch to learn more about other sessions that we have planned for you in the future. Thank you everyone.54:00