Transcript
Identity Security for AI Agents. Today when human workforce or employees join an organization, every organization has a defined structure, process and blueprint of how these employees get created in an HR system, what access do they get in their enterprise applications. Such a process, such a blueprint does not exist for AI agents. Everybody refers that AI agents should be treated as first class citizens, they should be treated as identities. What does that even mean? Let us take a look at in this Chalk Talk. Those organizations are building these agents, you are building them into your pro code, low code or no code platforms like Agent Core, Vertex AI from Google, Microsoft's Foundry or you can also have platforms like Crew AI, Lanchain and many more. You can build platforms and on SaaS platforms like Salesforce, AgentForce or ServiceNow Assist. Developers are also building platforms on their workstations or servers. So this is where the agents are getting built in an organization. And then these agents are also getting deployed on agent runtimes. For example, it could be Agent Core once again. It could be Bedrock. It could be Copilot. It could be compute layer like Kubernetes clusters, EC2 instances, virtual machines, servers. So, this is the landscape of how your agents are getting created by your developers, vibe coders, business users on different platforms. And then the same agents are also getting deployed on runtimes, where they are actually running into production and touching your enterprise applications. The question is, where do you and how do you create hooks so that an identity team knows that these agents are created and being run in your organization? And there are many different ways. The first one is you build what we call it as hooks. Second one is once you build these hooks, you push them into a central agent registry. And this could be pushed through hooks. These could be through APIs. These could be through events. There are many different ways in which these agents can be registered by understanding where the source of getting them built are and then pushing them onto agent registry. That is your step number one. And same thing, you do it even from your runtime environments. I am just creating another same version of agent registry. Here is the same one where you are getting it from your agent runtimes, again through APIs, understanding your logs, understanding your service accounts through which these agents are touching your systems or even sending this information as events. Now the agent registry knows where your agents are. Now what do we do about that? The first thing what we do here is we establish ownership of these agents. By ownership, what we mean is that you define who the business and technical owners of these agents are, number one. Number two, you also define the transfer ownership rules, which means as your owners are changing jobs, they are leaving their jobs, who becomes the next human counterpart who is accountable for what these agents are touching, doing and performing transactions on your enterprise applications are. And last but not the least in this case is you also define all your rules and policies which are being informed and enforced by the agent registry. Now this is your step one in your overall lifecycle process for your agents where you are issuing an identity to an agent, you are establishing ownership, you are establishing the policies of when and how the ownership agent of the agents are getting created and last but not the least the transfer rules as well. Once you have that, you send this information into a posture management layer. And the posture management layer as you would have seen in my previous chalk talk, you define the agents inventory, you also start defining the trust scoring for these agents. You collect all these different signals and you start defining are these agents trustworthy, you can look at signals like are they certified, what kind of systems are they accessing, who are the owners and many other different signals to start creating scores for your agents. You also look at things like what access paths these agents have and what are they doing with that access. So information about when your agents are getting registered and all of this feeds into the agent registry and then the agent registry starts feeding it into your posture management. So this becomes a continuous loop of how you are discovering agents, issuing identity, defining the owners and then feeding that into your continuous posture management. Once the posture management layer gets built, the next layer what you build is the governance layer. This is an intelligent and continuous governance layer which does couple of things. One it looks at all the different access and continuous reviews of what your agents are doing at any given point of time. Second you start looking at different SODs and if there are any separation of duties conflicts which are occurring because of agents transacting having access to different systems and should you be looking to clip, should you be looking at looking to clean up those kind of toxic combinations of your access. The third one also becomes very important is how do you start defining access recommendations for your agents. So agents having standing privileged access or agents how can they be requested by other agents or human counterparts for different type of transactions they are trying to do in your ecosystem. So that becomes the most important layer of your intelligence and governance layer on top of it. And needless to say anything and everything what we do always has an agentic workflow on top of it. This agentic workflow allows always any human counterpart or any agent to converse and have a very easy NLP based experience to talk to all this data what we are collecting. At any given point of time this is all about what an agent has experienced its life cycle from getting or issuing an identity to being managed on a continuous manner. The last and the most important aspect of this is about the killing of an agent or off boarding of an agent which also falls in the governance layer. It simply means that at any given point of time when an agent's job is done or if an agent has been marked as rogue and it should no longer exist in the system again the agent registry should fire off that information into posture, posture feeding it into the governance layer ensuring that your organizational security policies get enforced and such agents get removed from your ecosystem. Friends, I hope you like this Chalk Talk. This explains every step an organization have to think about right from the birth of an agent issuing an identity to an agent all the way to getting it registered on boarded managing the agents governance life cycle as well as off boarding an agent. It becomes extremely imperative for every organization to think about agents as first class citizens and ensuring that you have the same rigor what you have put in place as what you have for human identities. I am very much looking forward for you all to join me in my next Chalk Talk session which will be more about once the agents are created and running in your production environment how do you monitor, how do you ensure that they are doing exactly what they are supposed to be. They are accessing the same systems what they are supposed to at run time and that is where we have a very unique concept of Savings Access Gateway coming into picture. So, I will be very happy to share about how we are designing, how we are working with our design partners on that and I look forward to having you all there. Thank you once again for your time and I will see you there.
