Taesoo Kim in Tech Times: Microsoft MDASH Gains Defender Integration at Build 2026: 96.55% on CyberGym Benchmark

Microsoft's 100-agent AI vulnerability scanner entered its next phase at Microsoft Build 2026 on June 2, 2026, when the company opened an expanded preview of MDASH — the Microsoft Security multi-model agentic scanning harness — and plugged its findings directly into the Microsoft Defender Portal. Enterprise security teams that have struggled to keep pace with AI-accelerated attacks now have access to a system that moves beyond pattern-matching: MDASH traces full exploit chains across codebases, validates whether each flaw is genuinely exploitable, and surfaces results enriched with production risk signals — all without leaving the tools security operations teams already use.

The timing is deliberate. Microsoft's own Digital Defense Report 2025 documented threat actors increasingly using AI to accelerate vulnerability discovery and exploitation. Rule-based static analysis tools, which match code patterns against fixed rulesets, cannot reason across multiple files, ownership boundaries, or concurrent execution paths — exactly the conditions where the most dangerous bugs hide.
How MDASH Transforms AI Vulnerability Discovery

MDASH was built by Microsoft's Autonomous Code Security team, which drew several members from Team Atlanta — the Georgia Tech group led by Taesoo Kim that won the DARPA AI Cyber Challenge in 2024 and its $29.5 million prize. Kim, now Microsoft's VP of Agentic Security, describes the system's core architectural insight in one sentence: "The model is one input. The system is the product."