Microsoft announced new AI-agent security controls for Microsoft 365 E7 and Project Perception on 19 October, alongside an expansion in detection from 25 to 60 agent types, GeekWire reported.
The tools continuously assess each agent’s risk using signals that include the data it can access and whether its actions stray from its assigned task. If the assessed risk rises, they can withdraw an agent’s ability to make changes while preserving read access, require a person to approve a high-stakes action, or shut the agent down.
The expanded detection covers agents running on employees’ computers, including Anthropic’s Claude Code, OpenAI’s Codex and GitHub Copilot. Microsoft is also adding Guardian Agents to Project Perception. They can investigate a change in another agent’s risk and recommend or take containment steps. AI Affairs previously covered Microsoft’s introduction of ISOC in Defender for agentic security.
Microsoft launched the E7 licensing level in May. It costs $99 per user per month and includes Microsoft 365 Copilot, the E5 suite, Entra Suite and Agent 365.
