Cisco is buying Astrix Security, a startup focused on locking down AI agents and non-human identities, according to The Information. The deal puts one of the world’s largest networking and security vendors directly into the agent security business, a category that barely existed eighteen months ago.
The Information reports that Cisco is making the move as enterprises start deploying AI agents that act on their own behalf, hitting APIs, accessing data, and executing tasks without a human in the loop. Astrix’s pitch is straightforward: every agent, bot, and service account is an identity that needs to be tracked, scoped, and revoked just like a human employee.
What stands out here is the timing. Cisco isn’t waiting to see whether agent security becomes a real category. They’re betting it already is.
Why agent security suddenly matters
For years, identity security meant humans logging in with passwords and MFA. The new wave of AI deployments breaks that model. An agent built on top of GPT, Claude, or an open model can hold tokens, call tools, chain actions, and touch sensitive systems thousands of times a day.
A few things changed fast:
- Enterprises are wiring agents into Salesforce, ServiceNow, internal databases, and cloud consoles.
- Each agent typically holds long-lived API keys or OAuth tokens with broad scopes.
- Most security stacks were never built to inventory or govern these credentials.
- One leaked agent token can do more damage than a phished employee, because agents don’t sleep and don’t get suspicious.
Astrix built tooling to discover those agent identities, map what they can access, flag risky permissions, and shut them off when something looks off. That’s the gap Cisco is buying.
What Cisco gets out of it
Cisco’s security business has been on an acquisition tear. Splunk closed last year for $28 billion. Isovalent and Robust Intelligence followed. Astrix slots into that stack as the piece that watches what AI agents are doing once they’re deployed.
The play is bundling. Cisco can now pitch CISOs a single story:
- Splunk for the SOC and observability layer.
- Robust Intelligence for model-level threats and red teaming.
- Astrix for runtime agent and non-human identity governance.
- Cisco’s existing network and zero-trust products underneath all of it.
That’s a coherent answer to the question every security buyer is getting in 2026: “how are you securing the agents your engineering team is shipping?”
How this compares to the status quo
Until this deal, agent and non-human identity security was a fragmented startup market. Astrix, Oasis Security, Token, Andromeda, Clutch, and a handful of others were all chasing the same buyer with overlapping pitches. Most enterprises were either ignoring the problem or stitching together CSPM tools and homegrown scripts.
A Cisco acquisition consolidates the category in one move. Expect a few follow-ons:
- Palo Alto Networks and CrowdStrike will look at the remaining independents.
- Microsoft will likely lean harder on Entra to cover agent identities natively.
- Pure-play startups will need a clearer wedge or a faster path to revenue.
What practitioners should do now
If you’re running security or platform engineering at a company shipping AI agents, this deal is a signal to get moving on a few basics:
- Inventory every agent, bot, and service account that holds a token. You probably have more than you think.
- Scope tokens tightly. No agent should have admin-level access by default.
- Rotate credentials on a schedule and log every action an agent takes.
- Pick a governance tool now rather than waiting for the bundled enterprise pitch.
The broader read: AI security is splitting into model security, data security, and agent security, and each one is getting its own budget line. Cisco just planted a flag on the third.
More details on terms and timing are at the original report from The Information.