Microsoft MDASH Shows How Multi-Agent Security Systems Can Improve Vulnerability Discovery

Short summary

Microsoft's MDASH is a multi-model agentic scanning harness that coordinates multiple AI agents for vulnerability discovery, scoring 88.45% on the public CyberGym leaderboard. It found 16 CVEs during a Patch Tuesday evaluation and achieved 96-100% recall against historical MSRC cases for specific Windows components. The key insight is that orchestrating models, tools, and verification steps matters more than any single model's capability.

• •MDASH scored 88.45% on CyberGym and found 16 CVEs during Patch Tuesday evaluation
• •Multi-model agentic pipeline outperforms single-prompt approaches for security research
• •Governance questions around access controls, validation gates, and auditability are critical for enterprise adoption

Generated with AI, which can make mistakes.