Networking as AI Control Plane
Cisco has pushed into AI observability and identity with moves named Galileo and Astrix, positioning the network as a control plane for agentic workloads. The company framed these initiatives as part of a broader strategy to integrate networking, identity and observability for on‑premises and hybrid AI deployments. (networkworld.com, blogs.cisco.com)
Artificial intelligence agents need two things to work inside a company: a way to see what they are doing and a way to control what they can touch. Cisco spent the past week moving on both. (networkworld.com) On April 9, Cisco said it intended to acquire Galileo Technologies, an artificial intelligence observability company whose software measures model quality, safety, and reliability in production. Cisco said Galileo will extend Splunk’s observability portfolio, which it gained through its Splunk acquisition. (blogs.cisco.com, cisco.com) A few days later, reports said Cisco was in advanced talks to buy Astrix Security for about $250 million to $350 million. Astrix sells identity security for artificial intelligence agents and other non-human identities, such as service accounts, tokens, and machine credentials. (crn.com, astrix.security) Observability is the dashboard and flight recorder for an artificial intelligence system: it tracks prompts, outputs, failures, and drift after deployment. Identity is the badge system: it decides which agent can call a payroll application, open a customer database, or use a software tool. (blogs.cisco.com, cisco.com) Cisco is trying to tie those layers to the network itself, arguing that the network can become the place where policy, visibility, and enforcement meet. The company has been making that case across its AgenticOps, Hypershield, and Zero Trust for Agentic AI launches since 2025 and early 2026. (networkworld.com, networkworld.com, cisco.com) That pitch is aimed at companies running artificial intelligence in their own data centers or across hybrid setups, where cloud services, internal applications, and network controls all have to work together. Cisco’s customer-facing teams have been promoting on-premises artificial intelligence infrastructure as part of a broader push to keep sensitive workloads close to enterprise data. (blogs.cisco.com, blogs.cisco.com) Cisco has also been building the security case around agent software. In March, the company said 85% of enterprises were experimenting with artificial intelligence agents and launched Zero Trust for Agentic AI to connect identity, access control, and behavioral protections. (blogs.cisco.com, cisco.com) Galileo fits the “see everything” side of that strategy. Network World reported Galileo’s technology will be folded into Splunk so customers can monitor artificial intelligence agents, evaluate outputs, and spot failures or unsafe behavior in real time. (networkworld.com) Astrix fits the “trust nothing by default” side. Its platform inventories artificial intelligence agents, model context protocol servers, and other non-human identities, then flags over-privileged or unowned access that could let software act far beyond its intended role. (astrix.security, cloudsecurityalliance.org) Cisco is not alone in chasing this layer of the stack, and the Astrix deal had not been formally announced by Cisco as of April 15. But the pattern is clear: Cisco wants the network, plus identity and observability, to be the operating system for artificial intelligence inside the enterprise. (crn.com, networkworld.com)
