Cisco’s Astrix Deal Shows AI Agents Have an Identity Problem

AI agents are starting to cause a whole new set of problems for security teams, and it’s not just about what these agents are capable of. It’s also about proving who they are when they start moving around the company systems.

That’s where Cisco’s planned buyout of Astrix Security comes in. On May 4th 2026, Cisco announced that they’re planning to buy Astrix Security Ltd, a company that works on non-human identity security. Put simply, Astrix focuses on the non-human credentials, like API keys, service accounts, OAuth tokens, certificates, and secrets.

And you can’t just write these off as minor details. AI agents are relying on these credentials to connect to business tools, pull data, update systems, or carry out tasks for users.

For years, identity security has been all about people you log in, your device gets checked, you might get asked for 2FA, and your access can be revoked when needed. 

But AI agents don’t fit neatly into this picture. They might do work on behalf of a person, team, workflow, or app, and they can run in the background while swapping back and forth between systems, which makes the identity question way harder.

Now security teams have to start asking a new set of questions. Firstly, which agent is doing the work? What credentials is it using? What systems can they access? Is it given more access than necessary? And can anyone keep an eye on it to see if it’s being misused?

Why non-human identity is getting a whole lot tougher now

Non-human identities aren’t new—companies have been using them for years to let machines connect with each other. The problem is, AI agents can ramp up the places where these credentials are getting used.

This creates a visibility gap: the business team may understand what the agent is supposed to be doing, but the security team may not be aware of all the keys, tokens, or service accounts involved.

That’s when things start getting a bit dodgy. Some of these credentials get left active for too long, have more permissions than they actually need, or aren’t even properly documented. If one of them gets exposed, an attacker might be able to use the same access the agent has.

Why Cisco wants Astrix

Cisco isn’t just buying a new security product to plug in; it’s trying to gain more control over the identities of all the AI agents and machines that are quietly logging on behind the scenes.

Astrix has been working on this for years, and their technology is designed to assist businesses in determining who these non-human identities are, where they are appearing, and how to mitigate the risks associated with them.

Cisco will integrate Astrix into its Identity Intelligence, Secure Access, and Duo Identity and Access Management solutions. As a result, security teams will have a much better understanding of which agents are out there, what credentials they are using, and whether their access is still valid.

Then they want this data to feed into tools like Splunk. This is significant because when agents are active, their activity can be dispersed across various identity tools, SaaS platforms, cloud services, and internal apps. Security teams must be able to see the big picture rather than just a collection of fragmented signals.