Project Perception
Microsoft's agentic-insider defense: discovers 60 agent types, scores risk and uses Guardian Agents to supervise other agents.
Overview
Project Perception is Microsoft's security effort first aimed at AI-driven external attacks, which on October 19, 2026 was extended to treat AI agents already running inside companies as insiders. The update adds Guardian Agents, AI supervisors that check other agents and can limit permissions or block actions, continuous per-agent risk scoring that can block changes, require human intervention or shut an agent down, and discovery expanded from 25 to 60 agent types including Claude Code, OpenAI Codex and GitHub Copilot. Analysts note coverage is strongest for endpoint and in-house agents and weaker for agents bundled inside third-party applications, and that authorization remains the hardest open problem.
This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.
Key features
- Guardian Agents supervise other agents and can limit permissions or block actions
- Continuous per-agent risk scoring with block, human-review or shutdown responses
- Agent discovery expanded from 25 to 60 types, including Claude Code, Codex and GitHub Copilot
- Dynamic permissions for agents
- Announced alongside the Microsoft 365 E7 packaging per trade coverage
Use cases
- Inventorying coding and desktop agents running on corporate endpoints
- Containing over-permissioned or misbehaving agents
- Reporting agent risk to security and compliance teams
Pricing
unknown: trade coverage ties it to Microsoft 365 E7 but no standalone price is published in the sources reviewed.
Pricing reflects the source record and may change. Confirm current terms with the provider.
Dates and sources
- Released
- Not listed
- Record updated
- 26 Oct 2026
- Record last checked
- 26 Oct 2026