Microsoft AI Updates: October 20, 2026

1. Microsoft Adds Risk Scoring and Kill Switches for AI Agents in Microsoft 365 E7 and Project Perception

Microsoft. Microsoft announced a set of controls that treat AI agents as potential insider threats, positioned under Project Perception and Microsoft 365 E7 (which bundles Agent 365 with Microsoft 365 Copilot and lists at $99 per user per month). Dynamic Agent Risk gives each agent a continuously updated risk score from identity, endpoint, cloud, network and runtime signals; Dynamic Conditional Access extends Entra Conditional Access so rising risk can strip write access, require human approval, or shut the agent down; and Dynamic Permissions grants temporary, task-scoped access capped at the authority of the user the agent works for. Microsoft also introduced a Microsoft Execution Container sandbox with monitoring outside it, Guardian Agents in Project Perception that investigate risk changes and recommend or apply containment, and expanded local agent discovery to 60 agent types (including Claude Code, Codex and GitHub Copilot), up from 25. No availability dates or pricing for the new controls were given, with more detail promised at Ignite starting November 17. Source (SiliconANGLE)