AI & Enterprise
Microsoft says AI vulnerability discovery is moving beyond research, signals expansion
Microsoft on May 18 unveiled an agentic security analysis system called Multi-Model Agentic Scanning Harness, or MDASH, covering vulnerability discovery, verification, proof and response. The company said MDASH uses an ensemble architecture combining frontier and distilled models, with more than 100 specialised AI agents running end-to-end workflows. Microsoft said it identified 16 new vulnerabilities in Windows networking and authentication stacks and posted 88.45 percent on the CyberGym benchmark leaderboard.
