THE DAILY EDITION · 19 OCTOBER 2026
Microsoft says agent monitoring must become continuous
Why this story?
Microsoft is urging continuous monitoring of AI agents and more access to model reasoning, a concrete security proposal tied to its latest agentic-insider framing.
Microsoft says security systems should move from watching what AI agents do to examining why they do it, including whether a model’s reasoning fits the task it was given. The company said that would require more transparency into the model itself, not just input and output tokens, and said this would strengthen containment, accountability and traceability.
In the same post, Microsoft said monitoring and securing those agents must be continuous, with real-time understanding of what they are doing and what risks their actions create. It described agents as “agentic insiders” and said defenders would need “Guardian Agents” to watch and moderate every action before it happens. [P2
Microsoft also said an agent combines a model, an orchestration harness, tools and resources, and a runtime, and that securing them requires protection across identity, permissions, data, endpoints, network, cloud and runtime. The company linked that view to Project Perception, which it announced in July as an agentic security system using red, blue and green agents.