GitHub and Microsoft Cut 9,791 AI-Found Flaws to 99 Verified Reports
GitHub and Microsoft announced on November 17 the results of a pilot that uses AI to find vulnerabilities in open source software, then delivers only verified findings and proposed fixes to maintainers.
1 source 11m ago
Reporting from 1 source: GameBusiness.jp.
Why it matters
The pilot shows the bottleneck in AI vulnerability hunting has moved from finding bugs to verifying and fixing them, and GitHub's own admission that the verification step does not scale yet is the number that matters more than the 9,791.
Key Facts
- GitHub and Microsoft announced the pilot results on November 17.
- The pilot examined six projects: AutoGPT, Caddy, Express, FastMCP, ImageMagick and Netty.
- 9,791 detected issues were narrowed to 99 reports delivered to maintainers.
- As of November 15, 55 reports had been fixed and 14 fixes were already in releases.
- Verification took seven researchers about two days per project, and GitHub says the work does not scale yet.
Six projects were examined and 9,791 issues were detected, narrowed to 99 reports. As of November 15, 55 had been fixed and 14 of those fixes were already in releases.
The six projects were AutoGPT, Caddy, Express, FastMCP, ImageMagick and Netty, all participants in the GitHub Secure Open Source Fund, and maintainers consented in advance to the investigation. Caddy fixed all 22 reports it received within a week. ImageMagick, maintained by two people, also received 22 reports.
The pipeline runs in five stages: Microsoft's agent-based MDASH system finds issues and drafts fixes using the MAI Cyber model, a GitHub Security Lab AI triage agent removes duplicates and checks validity, researchers at Microsoft FORGE Lab manually verify what remains, reports go out through GitHub's private vulnerability reporting with a proof of concept and proposed fix, and maintainers decide what to fix and when. Verification took seven researchers about two days per project.
Written with AI by Yomimono from the 1 cited source below, including Japanese-language reporting where cited. See our editorial policy.
Add Yomimono as a preferred source on Google
Sources
- GameBusiness.jp AIが見つけた脆弱性9791件を99件に精査、修正方法も報告―GitHubとMicrosoftがOSS 6件で試験