Beyond the Benchmark: Advancing Security at AI Speed
Microsoft, Wednesday, June 17th, 2026
Microsoft details how its MDASH agentic AI vulnerability detection system reached 96.5% benchmark accuracy in production.
This Microsoft Security blog post describes the advancement of MDASH, a multi-model agentic scanning system that discovers, validates, and remediates software vulnerabilities at enterprise scale. The system has moved from research into production use across Windows, Azure, and identity systems, achieving 96.5% accuracy on industry benchmarks.
Recent results include discovery of multiple high-severity vulnerabilities in Windows components such as Hyper-V, DNS Client, and Active Directory Domain Services. The article analyzes performance gains and how newer AI models could push accuracy past 98%.