GitHub and Microsoft pilot Codename MDASH to deliver verified security fixes to open source maintainers
·2026.11.18 03:29
Key point
The pilot used Microsoft's Codename MDASH and human verification to filter thousands of AI findings into 99 private reports, with 55 fixes already implemented across six projects.
Details
- GitHub and Microsoft piloted Codename MDASH to shift the burden of security triage from maintainers to the reporting side, delivering verified findings with suggested fixes rather than raw scan results.
- The process filtered thousands of potential issues down to 99 private reports for six projects (AutoGPT, Caddy, Express, FastMCP, ImageMagick, Netty); as of November 15, 55 fixes were implemented and 14 were in release.
- Maintainers reported significantly reduced time-to-fix, with Caddy resolving all 22 reports in under a week and ImageMagick pushing a fix in 15 minutes, compared to the usual 30-60 minute average for security advisories.
- The pilot requires human verification by Microsoft FORGE Lab researchers (approx. two days per project) to ensure quality, a step that does not yet scale but establishes a standard for private, fix-inclusive reporting.
- GitHub is integrating these lessons into private vulnerability reporting and plans to offer AI security scans with pre-checked findings to all GitHub Secure Open Source Fund alumni, who also receive GitHub Copilot Max.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.