News
AI Agents as Insider Threats: Inside Microsoft's Bet on a Unified Security Stack
First reported darkreading.com
Page published
Earliest dated coverage: 23 Oct 2026 · First observed: 23 Oct 2026 · Latest dated coverage: 23 Oct 2026
Coverage timeline
Single-source incident — one report is available.
Why it matters
Microsoft’s Project Perception announcement highlights agent inventory, identity, permissions, and access revocation as controls for managing AI agents that can be manipulated or compromised.
Microsoft expanded Project Perception with capabilities to discover organizational AI agents, assign identities and permissions, assess risk, and revoke anomalous agents’ access, according to Dark Reading. The report describes Guardian Agents for monitoring and containment while highlighting concerns about using agents to police other agents.