Microsoft added tools to monitor AI agents as potential insider threats. The tools are part of Microsoft 365 E7 and Project Perception. They continuously assess each agent's risk. The system can block changes, require human approval, or shut the agent down. Detection now covers 60 agent types, up from 25. Guardian Agents can investigate risk changes and recommend containment. The features are in private preview. Broader availability is planned for Microsoft Ignite in November. E7 costs $99 per user per month. It includes Copilot, E5 tools, Entra Suite, and Agent 365.
What changed
Agents could act without continuous risk assessment or human oversight.
What it unlocks
Blocking an agent from making changes, requiring human approval, or shutting it down based on risk.
- 60 agent types now detectable
- 25 agent types previously detectable
- $99 per user per month for E7
- Nov 17-19 broader availability
What you need to act on it
- Microsoft 365 E7 plan
- Project Perception access
- private preview initially
Sources