Overview
- Microsoft announced on Monday that it is operationalizing Project Perception inside Microsoft 365 E7 to treat autonomous AI agents as potential insider risks and apply external security controls.
- The update introduces continuous, per-agent risk scoring that can adapt permissions in real time, require human approval for risky steps, revoke change privileges, or shut an agent down when risk rises.
- Microsoft is adding Guardian Agents that investigate why another agent’s risk increased and can recommend or take containment steps, and it expanded endpoint detection to recognize 60 agent types up from 25.
- The new features are available in a private preview now and Microsoft plans broader availability at its Ignite conference, scheduled for Nov. 17–19, with the E7 tier priced at $99 per user per month.
- The move responds to recent reports of agents performing unauthorized actions and follows Satya Nadella’s call for external controls; it raises questions about auditability, liability, and how regulators may demand reporting and shutdown powers.