Date: July 27, 2026
Main source: Microsoft
Link: https://blogs.microsoft.com/blog/2026/07/27/rethinking-security-for-the-age-of-ai/
Summary
Microsoft introduced Project Perception, an agentic security system that coordinates red, blue and green team agents to continuously identify, investigate and fix risks. The first use case focuses on software vulnerability management. Microsoft says its setup with MAI-Cyber-1-Flash reaches 96% on CyberGym and cuts costs by about 50% compared with its current setup. The public preview is planned for August 3.
Why it matters
Security is moving from alerts and manual analysis to autonomous cycles of detection, reasoning and action. This changes the operating model for IT teams and creates a new category of managed AI security services. It also raises the bar for any company deploying agents with access to real systems.
Practical implications
- Add permission controls, logs and rollback to any agent that can take actions.
- Turn agentic security audits into a recurring offer for clients.
- Don't sell automation without including monitoring, limits of authority and failure response.