Cisco's $400M Astrix bet: non-human identity is now a platform category

Cisco announced on Monday it has agreed to acquire Astrix Security, the Israeli startup focused on securing non-human identities — the API keys, service accounts, OAuth tokens, and machine credentials that applications and AI agents use to access enterprise systems. SecurityWeek’s Ryan Naraine reported the deal the same day. Calcalist reported the deal value at roughly $400 million; Cisco did not disclose financial terms in its own announcement. Astrix raised a $45 million Series B in late 2024, which puts the reported price at around nine times its last private valuation.

The rationale Cisco published on the same day is broadly the rationale every non-human identity vendor has been pitching for two years. Enterprises have lost track of their machine identities; they have an order of magnitude more of them than human ones; the credentials are over-privileged, long-lived, and mostly invisible to existing IAM stacks; and the rise of agentic AI — autonomous software that holds and exercises credentials — is going to make the problem worse, not better. Astrix’s product covers discovery, lifecycle management, and detection-and-remediation of over-privileged or compromised non-human accounts. Cisco’s stated plan is to fold those capabilities into its identity intelligence, secure access, and Duo IAM lines.

What the deal actually tells us is about market structure rather than technology. Non-human identity has been a category most CISOs treated as a backlog item. It is now consolidating into the identity platforms, and Cisco is the latest large vendor to move. Microsoft has been building NHI features into Entra. Okta has acquired and partnered. CrowdStrike has extended Falcon Identity to cover service accounts and tokens. Astrix is not the first NHI startup hoovered into a platform; it is the first to land at this kind of price tag, and the price tag is the bit worth noting. Nine-times-last-raise paid by a company that did not need to buy this capability — Cisco could plausibly have built it inside Duo — implies the platform vendors believe the category is going to be both large and contested.

For defenders, the practical question is not which vendor wins the NHI category. It is whether the enterprise has any policy for the non-human population at all. The honest answer in most organisations is no. Most identity programmes were built around the human population — joiner-mover-leaver, MFA, conditional access — and the non-human population was treated as plumbing the platform team handled. With agentic AI now consuming credentials at machine speed, plumbing has become a control surface.

The questions a CISO should be able to answer this quarter are practical and cheap to scope. How many non-human identities exist, in which systems, and who owns them? How many of the dormant ones still hold long-lived credentials that nobody is rotating? Which non-human accounts have privileges they did not have six months ago, and who approved the change? When an AI agent acts on behalf of a user, whose access is being exercised — the user’s, or a service account’s — and which one shows up in the audit trail when something goes wrong? These are not Cisco-specific or Astrix-specific questions. The Cisco-Astrix deal is a useful prompt for the inventory exercise that should already have been done.

A note on the disclosure itself. The Cisco blog and the SecurityWeek piece both lean heavily on the agentic-AI framing — autonomous AI agents holding credentials, acting on enterprise systems, generating their own access patterns. The framing is directionally correct but light on operational detail. There is no published claim about how Astrix’s detection layer behaves against an AI agent that uses its credentials normally for 27 days and exfiltrates on the 28th. There is no published quote from a customer running NHI controls in front of an agentic-AI workload at any scale. The product may be excellent against this exact threat model, but the announcement is a market-positioning artefact, not a capability benchmark. Buy on the demo, against your own data, with your own people running the workflow.

The Cisco-Astrix deal also lands inside a wider posture shift. Two months ago Anthropic’s Mythos and then Claude Security launches made the case that AI is going to compress the offensive timeline and that defenders need AI-native tooling to keep up. The platform vendors are now placing chips on the adjacent claim: that the AI workforce will consume identity at machine scale, and that the identity stack needs its own AI-native control plane. Whether that prediction lands is a 24-month question. The acquisition pace says the largest vendors are betting it does.