Updated
Updated · Microsoft · Jul 27
Microsoft Unveils Project Perception, Launching AI Security Preview on August 3
Updated
Updated · Microsoft · Jul 27

Microsoft Unveils Project Perception, Launching AI Security Preview on August 3

3 articles · Updated · Microsoft · Jul 27

Summary

  • Project Perception will enter public preview on August 3 as Microsoft’s new agentic security system built to detect, reason over and respond to threats at machine speed.
  • Microsoft says the platform is designed for AI-era attacks that generate exploits faster and scale campaigns further than human-led defenses can handle.
  • Three agent classes power the system: red team agents map attack paths, blue team agents investigate and prioritize risk, and green team agents take corrective actions to harden defenses.
  • A multi-model architecture underpins the product, with Microsoft’s MAI-Cyber-1-Flash scoring 96% on CyberGym—12 points above Mythos—and cutting software-vulnerability-management costs by nearly 50% in MDASH.
  • The launch reflects Microsoft’s push for a new 'Cyber Stack' that links signals, context, models and actuators across identities, endpoints, apps, data, clouds and AI systems.

Insights

Can Microsoft’s Project Perception really stop AI-powered attacks at machine speed, or does autonomous defense create a dangerous new attack surface?
Is Project Perception’s real breakthrough its cyber models, or the orchestration of red, blue, and green agents into one defense loop?
If Microsoft’s AI agents can find, judge, and fix vulnerabilities, what happens when their decisions are wrong in real enterprise systems?