微軟MDASH AI系統發現16個Windows漏洞
微軟部署了一個名為mdash的自主系統，包含超過100個ai代理，發現了windows中16個新漏洞，其中4個是關鍵漏洞。
microsoft revealed the mdash (multi-model agentic scanning harness) system, designed to autonomously audit proprietary codebases like windows and azure that are absent from public training data. the system orchestrates more than 100 specialized agents across a five-stage pipeline, using 'debater' agents to contest vulnerability findings and 'evidence leaders' to generate proof-of-concept exploits. this process uncovered 16 new cves affecting the windows networking stack, with four rated critical due to unauthenticated remote code execution risks. mdash achieved a score of 88.45% on the cybergym benchmark, leading the leaderboard, though microsoft withheld details about the specific underlying models used.