Microsoft pits more than 100 AI agents against each other to find Windows vulnerabilities

Microsoft unveiled MDASH, a multi-model security harness that coordinates 100+ specialized AI agents to scan code, debate exploitability, and validate findings.

The system uncovered 16 Windows vulnerabilities disclosed on May 12, 2026, including four critical flaws, among them remote code execution issues.

MDASH also scored the top result on the CyberGym benchmark, highlighting the strength of multi-agent AI over single-model approaches.

The findings underscore how autonomous AI could speed discovery of serious security bugs across widely used software.