Cisco plans to acquire Astrix Security

Cisco is buying a security startup because the next identity mess in the enterprise probably won’t be human. It’ll be API keys, service accounts, OAuth tokens, and now AI agents acting on someone’s behalf. That’s the gap Cisco is trying to close with its planned Astrix Security acquisition, announced May 4. The basic bet is simple — if software agents are going to read data, make decisions, and trigger actions, they need the same kind of controls companies built for people. (blogs.cisco.com) ### What exactly is Astrix? Astrix is a startup founded in 2021 that built tools for discovering and governing non-human identities — the machine credentials that let apps, services, and automations talk to each other. Think API keys, service accounts, secrets, and tokens. Those things are everywhere, (blogs.cisco.com)t they can access, or whether they should still exist. Astrix raised a $45 million Series B in December 2024, bringing total funding to $85 million. (astrix.security) ### Why is Cisco interested now? Because AI agents make the old problem bigger and weirder. A service account sitting quietly in the background is one thing. An agent that can chain tools together, pull sensitive data, and take actions at machine speed is another. Cisco framed this as the rise of (astrix.security) a chatbot wrapper. Cisco’s own AI Readiness Index says only 24% of organizations can control agent actions with proper guardrails and live monitoring, and only 31% feel fully capable of securing their AI systems. (blogs.cisco.com) ### What does Cisco want to do with it? Cisco says Astrix will be folded into its broader security stack, not left as a side product. The named targets are Cisco Identity Intelligence, Secure Access, Duo, and Splunk. That matters because Astrix by itself can help find and govern machine identities, bu(blogs.cisco.com)t together, the pitch is end-to-end coverage — discover the agent, verify what it is, limit what it can do, and watch for abuse in real time. (blogs.cisco.com) ### Why are non-human identities such a headache? Because they’re easy to create and hard to clean up. Humans usually go through HR systems, joiner-mover-leaver processes, and MFA prompts. Machine identities don’t. They get spun up by developers, embedded in workflows, copied into configs, and forgott(blogs.cisco.com) but “which agent used which token to do what, and was that action allowed?” That’s a different kind of identity problem — closer to continuous governance than simple login security. (astrix.security) ### Is this a new category or old IAM with new branding? A bit of both. The raw ingredients are old — secrets, tokens, service accounts. But AI agents change the stakes because they behave more like actors than plumbing. Astrix itself says the rise of agentic systems turned a neglected identity bl(astrix.security)y has already been building AI Defense, agent identity features in Duo and Secure Access, and more AI-focused monitoring around Splunk. Astrix fills the missing governance layer around the identities those systems rely on. (blogs.cisco.com) ### What’s the catch? Buying the platform is easier than making enterprises clean up their sprawl. Most companies already have years of accumulated machine credentials, inconsistent ownership, and overlapping tools. So the hard part is not just detection — it’s policy. Which agents get first-class ide(blogs.cisco.com)ically? Cisco is buying a strong answer to discovery and governance, but customers still have to do the organizational cleanup. That’s the unglamorous part. (blogs.cisco.com) ### Why does this matter beyond Cisco? Because this is what a real AI security market looks like once the demo phase ends. The conversation is shifting from “can we deploy agents?” to “how do we control them once they touch real systems?” Cisco is using M&A to say agent security is not a feature tucke(blogs.cisco.com) to make the same call — build, buy, or risk leaving a new attack surface unmanaged. (blogs.cisco.com) ### Bottom line? Cisco isn’t just buying a startup. It’s buying a claim about where enterprise security is headed — toward a world where every useful AI agent also has to be treated like a risky employee. (blogs.cisco.com)