NEWS BRIEF
What happened
Microsoft’s security leadership has argued that enterprise agents should be governed as active participants with identities, tools, and permissions. Its October post calls for continuous monitoring and controls across the model harness, runtime, identity, data, and infrastructure layers, including systems that moderate agent actions.
What the announcement does not establish
This is a vendor security position and architectural direction. It is not evidence that any one monitoring agent can prevent all unsafe behavior or replace deterministic access controls.
Primary source: Microsoft: original announcement (2026-10-19). Sources checked November 3, 2026.