Agents Are Your Newest Insiders. Security Must Move at Their Speed.
Microsoft, Monday, October 19th, 2026
Microsoft extends Project Perception with Guardian Agents to secure AI agents as a new class of insider risk.
Microsoft argues that AI agents with access to company data and systems create a new kind of insider risk because they can make mistakes or be manipulated or compromised.
Citing Satya Nadella, it says controls over what an agent can do must sit outside the model.
Microsoft is extending Project Perception to secure the agentic insider, with Guardian Agents and shared signals with Microsoft Agent 365.
The approach is to observe agents continuously and adjust their authority as risk changes.