Microsoft's MDASH AI System Discovers 16 Windows Vulnerabilities

Microsoft's newly developed MDASH (Multi-Model Agentic Scanning Harness) identified 16 vulnerabilities in Windows, including four critical remote code execution (RCE) flaws. The vulnerabilities were found in key components such as the TCP/IP stack and IKEv2 service. These findings were part of Microsoft's May Patch Tuesday, which addressed a total of 120 CVEs. MDASH utilizes over 100 specialized AI agents to enhance vulnerability detection, achieving an 88.45% score on the CyberGym benchmark. The system is currently in a private preview phase with select enterprise customers. The vulnerabilities include CVE-2026-40361 and CVE-2026-40364, published on May 12, 2026. Microsoft emphasizes that the AI system's architecture allows for efficient vulnerability discovery at scale, marking a significant advancement in cybersecurity tools.

Key Points: • MDASH discovered 16 vulnerabilities, including four critical RCE flaws in Windows components. • The system achieved an 88.45% score on the CyberGym benchmark, outperforming competitors. • Microsoft's May Patch Tuesday addressed a total of 120 CVEs, with MDASH contributing significantly.