Defense at AI speed: Microsoft’s new multi-model agentic security system tops leading industry benchmark

Microsoft describes MDASH, a multi-model, agentic AI scanning harness that found and validated 16 Windows vulnerabilities (including several critical remote code execution issues) across tcpip.sys, ikeext.dll, netlogon.dll, dnsapi.dll and others; the post explains MDASH’s prepare/scan/validate/dedup/prove pipeline, provides benchmark results (100%/96% recall on historical MSRC cases for tcpip.sys/clfs.sys and 88.45% on the CyberGym benchmark), and gives technical deep dives into two notable CVEs demonstrating cross-file and concurrency exploitation paths.