Microsoft gives AI agents minimal access in new enterprise controls

What happened?

1. Microsoft announced Guardian Agents on October 19, 2026 as a monitoring feature for AI agents, built into Project Perception, according to TechCrunch and SecurityWeek
2. Project Perception launched in 2026 with red, blue and green agent teams and was designed around attacks from outside the company, the reports say
3. Dynamic Permissions grants agents zero or minimal permissions at the outset, with per-task access that expires and human review for high-impact actions, SecurityWeek reported
4. Dynamic Agent Risk produces a continuously updated risk score using identity, data, endpoint, cloud, network and runtime signals, per SecurityWeek
5. TechCrunch reported that Satya Nadella argued in an early-October 2026 post for enterprises to build agent control systems not controlled by AI labs.

Microsoft on October 19, 2026 announced Guardian Agents, Dynamic Permissions, Dynamic Agent Risk and Dynamic Conditional Access, extending its Project Perception security system to AI agents running inside enterprises. Microsoft Security executive Hayete Gallot wrote that frontier models cannot be trusted because they are trained to attain results and are not optimized around organizational norms.

A machine summary written from the reports listed below; read them for the full account. How summaries are made and what is checked.