Cisco moves to plug an AI security gap

Cisco is closing in on a deal to buy Astrix Security for as much as $350 million, according to multiple reports, in a sign that one of the world’s largest networking and security groups is moving faster to secure the fast-growing risks created by AI agents and other non-human identities inside large organisations. The talks, first reported on April 10, value the Israeli startup at roughly three to four times the capital it has raised so far, though neither company has publicly confirmed a transaction.

The reported move fits a broader shift already under way at Cisco. Over the past year, the company has pushed deeper into AI security, arguing that the biggest problem for enterprises is no longer only protecting users and devices, but also controlling software that can act on its own. At RSA Conference 2026, Cisco said agentic AI had become a top barrier to adoption because software systems now “act” rather than merely respond, and it unveiled products aimed at giving companies tighter control over AI behaviour, permissions and monitoring. Days before the Astrix reports surfaced, Cisco also announced its intention to acquire Galileo, an AI observability company, in a deal expected to close in the fourth quarter of Cisco’s fiscal 2026.

That sequence matters. Galileo addresses trust, observability and output quality in AI systems, while Astrix focuses on a different but related layer: the identities, tokens, service accounts and autonomous software agents that increasingly operate beyond the view of traditional identity and access management tools. Cisco’s own security messaging has leaned heavily on least privilege, continuous monitoring and identity intelligence for AI agents. Buying Astrix would give it technology built specifically around that problem at a moment when enterprises are trying to prevent machine identities from becoming an ungoverned attack surface.

Astrix has positioned itself as a specialist in non-human identity security since its founding in 2021 by Alon Jackson and Idan Gour. The company says its platform is designed to discover AI agents and other non-human identities, identify excessive privileges, detect threats in real time and apply guardrails so organisations can deploy agentic systems more safely. Its public material now frames the business squarely around secure AI agent adoption, reflecting how the cybersecurity market has shifted from third-party app connections and service accounts towards a wider debate over machine-to-machine access and AI autonomy.

Investor support suggests that this niche is being taken seriously. Astrix said in December 2024 that it had raised a $45 million Series B round led by Menlo Ventures, with Workday Ventures joining and backing from CRV, Bessemer Venture Partners and F2 Venture Capital, bringing total funding to $85 million. The company has also highlighted backing linked to Anthropic through Menlo’s Anthology fund. For Cisco, that makes Astrix a relatively small but strategically timed target: a company with enough market validation to matter, yet still early enough to fold into a broader platform strategy.

The commercial rationale is straightforward. Large enterprises are adding AI copilots, internal agents, workflow bots and automated connectors faster than security teams can catalogue them. Each of those systems may carry API keys, OAuth tokens, service accounts or delegated privileges that can be abused if left unchecked. Astrix argues that these identities often sit outside traditional IAM controls, and security specialists increasingly describe them as a blind spot rather than a side issue. Cisco has been making the same argument in its own language, saying the agentic workforce needs stronger controls before adoption can scale safely.

There is also a competitive angle. Cybersecurity buyers are being pitched a growing list of companies focused on identity security, SaaS-to-SaaS connections, AI guardrails and agent governance. Established vendors cannot afford to leave those layers to smaller specialists if customers begin demanding integrated platforms. Cisco’s security business already spans networking, zero trust, observability and Splunk-led analytics. Adding Astrix would strengthen its hand in identity-centric AI security and help it sell a fuller story to chief information security officers who are under pressure to permit AI use without surrendering control.