Microsoft Agent 365 is live: AI agents get a control plane

Microsoft is making Agent 365 generally available on May 1, 2026. The product was previewed earlier, but today matters because Microsoft is turning AI agents into governed enterprise objects, not just loose automations inside Copilot Studio, Foundry, or third-party tools.

The primary sources are Microsoft’s Agent 365 product page and the Microsoft Security blog post “Secure agentic AI for your Frontier Transformation.” VentureBeat also covers the launch today with the same core point: Agent 365 is a control plane for visibility, identity, access, policy, logging, and security around agents.

The practical shift is straightforward. Once agents can read documents, call APIs, update cases, and retrieve data on behalf of employees, they need to be treated more like digital identities than chatbot features. Microsoft connects Agent 365 to Entra, Defender, and Purview. That means agent inventory, Agent ID, conditional access, risk signals, DLP, audit, eDiscovery, and the ability to block risky agents.

The pricing is also a signal. Microsoft’s Security blog lists Agent 365 as a standalone product at $15 per user per month. VentureBeat reports that Microsoft 365 E7 bundles E5, Copilot, Agent 365, and advanced security at $99 per user per month. Microsoft is not only licensing the humans who use AI, it is pricing the governance layer around the agents they create.

For CIOs and CISOs, this is an early marker for where enterprise AI is heading. The next wave is less about “which model is best” and more about inventory, ownership, access, logging, and incident handling for autonomous systems. If teams are already building agents, the first questions should be: where is the agent registry, who owns each agent, what data can it reach, and how do we shut it down?

Assessment: Agent 365 is not magic, and it will not solve agent risk by itself. But Microsoft is making the governance layer concrete enough for leaders to demand it. Agent projects without identity, logging, and lifecycle management should no longer be treated as experiments. They are shadow IT with better demos.