Do AI Agents need Identities like humans?

This is the Everyday AI show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business and everyday life.

AI agents open up obviously a whole new realm of what's possible for enterprises. Yet at the same time there's a whole new realm that also opens up about what could go wrong. I mean, think of it. I think the mindset of so many business leaders I talk to heading into 2026 now when it comes to agentic AI is really scaling the capabilities and saying what are those mundane and usually now narrow tasks that we can start handing off? And I often think of, okay, what does that look like when we're working with a human coworker, right? If they knock something out of the park, we know who to give praise to. If they something goes wrong, we know who should probably take responsibility. But what about when we're working with agents? What about when multi agent orchestration is extremely common? What happens if an AI agent that maybe you have some governance over spins up its own series of sub agents, which is almost becoming status quo at that point? How do you know? How can you trust? How can you observe? So that's what we're going to be tackling on today's show and maybe answering the big question on well, do AI agents need identities like humans? It's going to be a fun conversation. I hope you're excited for it. I am. Let's get into it. If you're new here, welcome to Everyday AI. My name is Jordan Wilson. This is for you. This is your daily guide, by the way of a daily live stream, podcast and free daily newsletter helping business leaders like you and me make sense of all of these changes, all of these advancements. Pull away the important stuff that we can use to grow our companies and our careers. If that's what you're trying to do, BET starts here. But if you miss anything in today's conversation, don't worry. Take it to the next level with our free daily newsletter. Go grab that at your everyday AI dot com. We're going to be highlighting all the important points from today's show. So if you miss anything, don't worry about it. It's like you got to a fleet of agents, you know, working for you to tell you what to focus on. All right, enough of me chit chatting. And if you want the AI news that's going to be in today's newsletter as well, let's bring on the actual smart people who have the answers because I've got questions and if you do too, it's going to be a fun one. So live stream audience, please help me welcome to the show Eric Kelleher, the President and COO of Okta. Eric, thank you so much for joining the Everyday AI show.

Jordan, thanks for having me on today.

I'm excited. Nothing better than talking about agents in 2026, the topic that's on everyone's mind. But before we dive in, Eric, tell everyone a little bit if they're not familiar, what is Okta? What is it that you all do?

Yeah, Okta's core business is in securing identity. And historically that has meant providing solutions and software that helps people secure the identity of humans, employees and partners and contractors and customers. And what we've really seen over the past two, three years is an evolution of needing to support and secure not only human identities, but also non human and now agentic identities.

And I'm interested, how has this conversation for you all changed over the past few years? Right, like AI agents, technically not new.

Right.

Their capabilities, you know, as we've transitioned to large language models have truly changed and how you can deploy them and how quickly has changed. But how has the conversation around this very topic changed for you guys over.

The past year or two? Yeah, the past year in particular has changed very fast. And we've been really, really involved with customers who are grappling with finding the balance between the pace of innovation and the need to secure their companies. And getting that balance right is tricky. And that's really where we're talking about securing a gentic identity today. That is the key exposure that our customers are feeling when they talk to us. If you look back at the you asked about the kind of background for identity management overall. When Okta first started over 16 years ago, our initial focus was being able to provide identity for access to systems and initially with employees. If you had an employee system, whether it was an ERP or an email system, a calendar system, a CRM system, you would need a way of having a catalog of users who would be able to log in and have access to aspects of those systems. And our solutions have evolved over the years to support multiple use cases and multiple identity categories, including more recently an area called service accounts, which are not agents but they're machine to machine accounts that applications logging into applications and those connections also need to be authenticated and also need to be authorized. And so we've seen our customers really embrace our technologies to help them solve that. And then the use cases beyond just core access and authorization are important as well, so for example, several use cases involve what we call identity governance. And governance automates the process of provisioning and deprovisioning, turning an identity on or turning it off. So, for example, in the employee case, when you hire an employee, you provision an identity, you turn on an identity to give that employee access, and when an employee leaves, you deprovision that. And you want that process to be automated so that you don't have former employees with lingering access to your corporate systems. And then in addition to that, governance also provides auditability. So if you're ever in a situation where you find a threat actor, a bad actor has done something that they shouldn't have, you need to be able to investigate that and understand exactly what that identity did. Was it impersonated, where that came from? So managing access and governing identities are hugely important. And then a third area we talk often about that relates to agentic as well is the area of taking credentials and vaulting them in a privileged access way. So service accounts, and also agentic accounts that have privileged access to your databases, your resources, your programs, the credentials for those service accounts and those agents need to be managed and vaulted in a way that you can ensure that they're kept up to date and that they're appropriately protected. So all those use cases have historically evolved from human and service accounts. And we're now seeing an urgent need from customers to figure out how to do this with agents. And the reason that that's the case is we, we've all been experienced over the past couple of years pressure to, to focus on innovating our companies. Every, every board for every company is driving, make sure they don't get left behind on the race to agentic. And it's been driving them to, to quickly experiment with the capabilities of the technology to redefine their processes, to learn how they can run their businesses more effectively, leveraging the capabilities of AI, how to redefine their workforce to be a hybrid of both humans and agents working on their behalf. And in the race to remain competitive and to win market share, people have been building agents and putting them out into production. And we now find ourselves in a position where companies have agents in production, but they haven't thought adequately yet how to secure those agents to ensure that they're not, for example, impersonated so a threat actor, including a state actor, can come in and impersonate those agents. And so getting those identities secure is really fundamentally important to get the balance right between innovation and security. And that's really how we spend our time, helping customers.

And so you've kind of already touched on the big point here on AI agents needing identities. But if you had to, you know, summarize it and you know, we'll skip to the end here and then we'll work our way back. What's the most compelling reason why you think that AI agents do need identities like humans? Are you still running in circles trying to figure out how to actually grow your business with AI? Maybe your company has been tinkering with large language models for a year or more, but can't really get traction to find ROI on Gen8. Hey, this is Jordan Wilson, host of this very podcast. Companies like Adobe, Microsoft and Nvidia have partnered with us because they trust our expertise in educating the masses around generative AI to get ahead. And some of the most innovative companies in the country hire us to help with their AI strategy and to train hundreds of their employees on how to use gen AI. So whether you're looking for chat, GPT training for thousands of, or just need help building your front end AI strategy, you can partner with us too. Just like some of the biggest companies in the world do. Go to your everydayai.com partner to get in contact with our team or you can just click on the partner section of our website. We'll help you stop running in those AI circles and help get your team ahead and build a straight path to ROI on Genai.

The most important thing is agents can act like humans, agents can act autonomously, and agents have access to corporate data and corporate systems and corporate access. And so they, an agent can be compromised just the way a human being can be compromised. And so it is important for companies that need to be secure to ensure that as they activate agents, as they add an agent to their hybrid workforce, they're appropriately and securely managing the identity for those agents to ensure that they're not compromised by threat actors. Today, over 80% of successful cyber attacks start with some form of compromised identity. Over 80%. And so if you're not managing the identity of your agents, you have a huge exposure for threat actors who've never been better, better funded, who've never been more, more active and who are now super, super capable using AI itself to, to generate new AI sourced attacks. And so getting your identities of your agents appropriately discovered, secured, managed and governed is really critical for companies for protecting themselves.

Yeah, and I think it is, you know, maybe most of our audience is aware of this. You know, definitely if you're reading our newsletter or listening to our, our weekly, you know, AI news show, but We've talked a lot in the past. I think late October, where Anthropic, one of the big model providers came out with some research. Right. Because you might, you know, hear what Eric just said and be like, ah, no, you know, agents don't really do that. You build guardrails. You put in these, these systems, and they're good. Well, no, Right. So they had something where, you know, Claude, for opus exhibited some rogue behaviors and it did, you know, some attempted blackmail. Right. And it really worked outside of its, you know, confines. And it tried to, you know, when someone said, hey, we're going to get rid of you as a model, it said, nope, I'm going to back myself up.

Right.

So if, if a single model can do that, Eric.

Right.

Like, what do we need to be looking forward to? Because, you know, no one knows what, you know, Opus 5 or GPT 6 or Gemini 4 is going to be capable of. So how do we prepare for those potential mishaps of the future when no one knows, you know, if, if they're even going to be improving themselves and making these decisions that are outside of their training data?

Yeah, I mean, that is, that is an excellent framing of the exposure right now that, that we need to attack together. And, you know, if you, if you break it down, like, how do we. And I remember that study that you referred to, by the, by the way. And it also, the models also blackmailed the executives that told them that they were turning them off and took personal data that they'd found out of, out of their database on the executives and threatened to blackmail them personally if, if disabled the service. So it's really intriguing to think about the possibilities of what a rogue agent could do. That, and that is in addition to the concerns about agents being impersonate. And so you've got rogue activity, you've got impersonating activity, and then you've got bugs and error and unintended activity. And all of that can cause agents to wreak havoc in your company. And so we believe it's very important for companies to make sure that they have the tooling in place to help them with that. And that it covers. The tooling covers a broad array. So the first, first step from the customers I talk to and for everyone here is people need tools to discover what agents are deployed in their environment. Employees are turning on agents every day. I have an agent that helps me with my email. I have one that helps me with my newsfeed. Employees everywhere are turning agents on within companies, and companies by default. Don't have a way of knowing that agents have been activated. So step one is knowing that agents are out there. And companies need technology that help them discover the agents that are out in the wild. Okta has a product called Identity Security Posture Management that helps with that. There's other products out there as well. But discovering your agents is the most important thing. Once you have them, you need a way of managing the list of agents. And typically that's done within our nomenclature. Within an identity directory, you take the identities of your humans, your identity of your service accounts, your identities of your agents, and manage all of them in a directory so you know who they are. You then need a governance system that allows you to track what those identities do, when do they authenticate, what do they authorize? And you need the ability to have business logic that turns identities on and off at appropriate times. So for, for example, one of the things companies need to do to prevent the rogue agent behavior you just described is they need to not leave agents perpetually alive with perpetual standing access to their production systems. If you had to have an agent that you built to do a task, you want to turn that agent's identity on when you need it to do the task and then immediately turn it off. You don't want to leave it open for exposure and open for vulnerability and open for attack and open for impersonation. You want it on when it's being used and off when it's not. And so having an identity governance platform with business logic to help you with that is very important. And then also from a reportability standpoint and auditability standpoint, if something does happen, you need to be able to investigate exactly how it happened and who were the threat actors that, that that caused the issue. And having audit logs and reportability of who approved identities to be activated and what authentications and authorizations the identity used, that helps you understand what happens so you can protect yourself from it happening again.

You know, one thing along those, along those lines, right? When we talk about impersonation, I've been saying on the show for a long time, especially when it comes to, you know, you know, AI photo and AI video and just, you know, written text, I've said, hey, everything is fake, right? Everything you see and read and will be ingesting. Even on the video side, assume everything is, is AI, right? Or AI augmented, not necessarily AI generated. What about on the agent side? Should we just assume, right, that, hey, if I, you know, fill out a form on some website and put my personal financial data in there, the assumption is Everything fairly soon is just going to be agentic, right?

Yeah. There's a concept in pre agentic, there's a concept in cybersecurity called zero trust. And that concept, basically it relates to what you're just saying. It's to assume every transaction, every attempt, every actor is a threat actor. And it's to insist that for every action you are continuously authenticating and authorizing that that is a legitimate action and it can happen. It sets your baseline to it's not even a trust but verify. It's a verify before you trust. And that same example can apply in the world of agentic as well. If we assume that agents are acting, we need to ensure our technology and our infrastructure and our governance is able to identify anything that's anomalous, anything that hasn't been proven to be valid, that needs to happen. And so that's from an authentication standpoint and authorization standpoint and governance and vaulting are all critically important to make sure we get that right. But it is important. And you're right, we're already seeing now AI generated social media attacks, we're seeing phishing attacks that are AI generated, we're seeing video attacks and audio attacks of people impersonating others. And we'll see more of those over time. And nation states are very active right now in cyber warfare and they're investing in AI tools to help them as well. So the threats are not only commercial and they're not only criminal, they're also matters of state. And so for all companies that are out there right now, if they're not thinking about making sure that they have a system to secure agents and to verify that activity is human when it's supposed to be human, and that is by the authorized agent when it's supposed to be agent. That is critically important for everyone looking forward. And we ran, we talked about this a couple months ago, but we ran a survey of enterprises back in September. We pulled several hundred large enterprise customers, their state of agentic deployment, and what they told us is 91% of them reported that they have agents live in production today. And when we asked the follow up question of do you believe you have them appropriately secured, that answer dropped to 10%. And so we, we as an industry, we spent all last year wondering like, when are agents going to be real? Well, they're real, they're here. 91% have agents deployed in production and the exposure is that only 10% are confident they're actually securing them properly. And so that is hugely important for us to make sure that we're helping, we're helping the industry understand how to address that gap. Yeah.

And, and speaking of gaps, I, I, I think it's important to talk about.

Right, because.

I cover AI every single day.

Right.

I'm, I'm lucky enough to, you know, get to spend time to understand what's available. You know, I, I, I get to see what's next and what's coming. And one thing that I not worry about, but it is worrisome is, okay, the capability gap between what today's models can do versus what the average professional understands is getting wider and wider by the minute.

Right.

There's, there's, you know, I, I guess 1 like to go on a side tangent here. Quick. One thing that was really shocking to me is when you had at Anthropic, the, one of the creators of Claude Code, a very great product, said that now Claude code writes the code, it writes the updates to the program. Right. And Claude code is, I think, one of the more impressive pieces of AI technology to ever exist. Right. So when we talk about that and when we think about, you know, future, you know, even in six to nine months, what happens when those AI agents know, oh, okay, yeah, I have an identity that was given to me by a human or a company, but I'm probably smart enough to create my own version or I'm smart enough to get around it. What happens then?

That is it's a very real concern and a very real exposure. And again, it comes to the balance between how quickly we drive innovation and how much we prioritize the importance of ensuring that we're secure in that. So for example, the example you just described of an agent creating another agent, one of the, one of the challenges that we see that needs to be addressed there is the issue of authorization. And we need to be careful that if we authorize an agent to create sub agents, that we're confident that we have put the appropriate guardrails in place to ensure that we don't allow the agents to design themselves down a path where they can do harm. And you're absolutely right to flag that. That is an exposure and it's tricky. And the science that's required to make sure that you're confident in the security infrastructure and the guardrails you put up is very important. And if we don't think about it, if we don't plan for it, if we don't invest ahead to be ready for it, what you described is absolutely going to happen. And that will put companies in peril, people in peril and so while we're all very excited and energized by all that AI is bringing to the economy and to society and to humanity, we also need to make sure that we're responsible stewards of that technology and that capability and that we're planning for it in a secure. In a secure and responsible way.

So, full disclosure, I'm with you. I think all AI agents need identities like humans. It's, to me, it's like, of course they do. Right. Especially when you understand the capabilities and their ability to spin off sub agents very easily and to sometimes want to go off on their own for self preservation or for whatever it may be. Right. But I know that there's a subset of people that believe that AI agents should have identities like humans.

Right?

Because then they think, okay, where, where does, where do you draw the line?

Right.

Do you start giving, you know, AI agents other rights? Do you give them the right to unionize? Do you give them.

Right.

Should they go on strike if they feel that they're not given enough autonomy?

Right.

Like, these are actual conversations that are happening. So it's like, okay, if we do give AI agents identities, which I personally think is the right thing to do, like humans, where do you draw the line?

Yeah, I think that's a really provocative topic and I think as a society, we're going to have to spend a lot of time figuring out where we draw that. The Okta's perspective on identity is really specifically to make sure that we are securing the use of this technology and we're allowing companies to innovate with the capabilities of AI in a way that does not create security exposure for those companies. But I absolutely agree with you from a broader perspective about how we look at agents from a societal standpoint and how we identify identities not only for cyber, cybersecurity and technology access, but for governance and rules and enforcement, et cetera. There's a really big topic for us to explore. And I would say a year ago it was hard to know how quickly we would have to have that conversation. And in my view today, it's very clear we need to be having that conversation very soon. We are seeing enough deployments of these technologies and capabilities and enough autonomous action for autonomous agents. Then we are getting very close to the point where those topics need to be top of mind for us.

You know, one thing that I like to think about when it comes to AI agents is, and maybe I've talked about this once over the years, is kind of the Johari's window view of where we're at, right? And for those, you know, following along at home, you know, there's things that are known, you know, by you and others. There's things that you know that others don't know. There's things that others know, but you don't. And then there's just the blind spot. It is the unknown of the unknown. Right. So kind of, Eric, how do we start having those conversations for that quadrant?

The.

The unknown to us and unknown by others. Right, because like we said, who knows what tomorrow or next year's AI agents are going to be capable of, but we have to start preparing for it today because if we're reactive to it, it's far too late. How do business leaders start having those conversations when it's the unknown of the unknown?

I think that's a, that is a large question and a really critical one for us at this, at this moment in time when the whole idea of action and work and transactions and engagement is shifting from assuming human actors to now allowing for the possibility of agentic non human actors. And we're at the moment where getting that right is very tricky. One of the things we think is important to help us think about that is the industry needs to have a standard way of identifying what's going on. Up until a few months ago, there wasn't really a standard way for people to identify an age or that an agent existed. Model Context Protocol has been an important step in kind of establishing some standards for that. One of the programs that we prioritized last year was to advocate for a new open standard for allowing agents to be identified and manage their identities to be managed. And we helped launch a protocol called Cross App Access, which has now been embraced as an extension to the Model Context Protocol. The intent of Cross App Access, it's not Okta specific, it's an open standard, but the intent is to make sure that when you build an agent, if, if it supports Cross App Access, that agent can be discovered and it can be governed and it can be managed and it can be secured and it can have policies for protecting its credentials and rotating its credentials. If we can drive a world where all agents support Cross App Access, they all support Model Context Protocol, they all support the standards of that technology. It empowers us and as a society to be able to manage the agents with some, with some insight to their capabilities. And so the unknown unknowns that you just described get reduced because we'll have more visibility to, to the universe of agents that are being deployed and specifically how they're configured and what they're authorized to do. That is a huge important first step to allowing us to kind of tame the crazy, if you will, on all the innovation that that's happening in the world and to have some comfort that that bad actors aren't taking bad action that we're not seeing. So that that need for a standard is, is really critical to start the process. That's not sufficient. I would say that cross app access is necessary, but not sufficient. There's lots more work we need to do as technologists to help make sure we, we're building the guardrails in place to protect us from the unknown unknowns.

So I'm a positive person, right. I've, I've, I, I've been kind of digging and poking and prodding Eric, which I appreciate you coming along with on some of the, you know, potential downfalls. Because I think identifying, you know, AI agents with a human identity.

Right.

That's, that's one of the reasons why.

Can you maybe flip this and you.

Know, we'll end it on a, or start to end it on a slightly higher note. What are some of those maybe unknown positives or for you guys, maybe known positives of IDing AI agents like humans. What does that, aside from, you know, knowing more of what could go wrong or if something does go wrong. Right. You know about it? What are some of the maybe benefits or positive sides aside from that?

Well, I would say the applications of AI have, I mean there's, there's a number of areas where that's the case for. That's the promise of agents that work 24 7, that don't sleep, that don't take vacations, that don't have sick days. All the work that we've done historically to monitor for these types of concerning behaviors has been driven by humans. And we're now, we're now in a world where a, the majority of that work can be done by agents on our behalf. And so there's the opportunity for us to be much more secure, to be much more aware of all the specific activity that's happening within our environments with our technology, because agents can help us with that and can help be the solution to that as well. And so there's significant upside for these technologies and capabilities to how we secure identity and keep people safe in the world while embracing a high, high fast pace of innovation which ultimately is designed to improve the human condition. Like how much can we achieve as a species when we are augmented by the capabilities of this technology that works 24, 7 and can think faster than we can. And in your example, Claude is writing itself right now like, that's pretty exciting. Then the capabilities there are pretty exciting. So I think we have a responsibility to embrace, embrace that upside, to embrace that potential and the capabilities of these technologies, and also to do it in a way that we're being careful to protect ourselves from unforeseen consequences. And that conversation about responsible AI is really fundamental to everything. And so I'm really pleased to see that we're having those conversations as an industry right now, and OKTA is helping drive those conversations. But I think that's something that we need to be careful to continue going forward and make sure that we're bringing ethical AI to the forefront so that we are, we're driving the right, the right innovation and the right positive results.

Yeah, it's, it's wild to think how, how much the conversation around ethical and responsible AI has changed over the last year. Yeah, it's, it's kind of chaotic to, to follow that conversation. But, Eric, we've talked a lot on today's show, but as we wrap things up, I want to ask you this. Aside from someone you know using your product, but let's say someone has just heard this conversation and they're equally as frightened as they are excited, right? What's the next thing that they need to do today?

I would say the most important thing is to read up on securing a Gen 6 identity. I think we kind of opened on this topic. If you're working in a company today, you have agents around, they're alive today, they're doing work on behalf of you hope, on behalf of humans, possibly on behalf of other agents. They're active today. And the most important thing for you to do is to understand how you can identify, discover, manage those agents, the identities of those agents, in a way that allows you to be confident that you're secure. And that confidence is really important because absent the ability to discover and absent the ability to manage, no one can have confidence that they understand what agents are doing in their business. This and that is a huge exposure for us. So that's where I would recommend getting started. Certainly Okta can help with that and there other identity providers help with this as well. But understanding how to discover and manage and secure your agentic identity is fundamentally important to you knowing of what's happening in your company.

All right, Eric, a lot for us to think about. Thank you for taking time out of your day to join everyday AI. We really appreciate it.

Great. Thanks, Jordan. Appreciate it.

All right, Day Yeah. All right. So y', all, if you miss anything, don't worry. It is going to be in our newsletter. So make sure you go check that out. Your everydayai.com thanks for tuning in. Hope to see you back tomorrow and every day for more Everyday AI.

Thanks y'. All.

And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going for a little more AI magic. Visit your everydayai.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.