GitHub and Microsoft pilot AI that finds and fixes 55 open source bugs

Covered by 2 sources
Read onGitHub Blog (AI & ML)· 11 min
GitHub and Microsoft tested Codename MDASH, an AI system that detects and suggests fixes for security vulnerabilities in open source projects. The pilot filtered nearly 10,000 potential issues to 99 verified reports sent privately to maintainers, who fixed 55 bugs and publicly released 14 patches. This shows AI can handle detection, but maintainers still play a key role in applying fixes.

What happened

GitHub and Microsoft conducted a pilot using an AI system called Codename MDASH to identify and suggest fixes for security vulnerabilities in six open source projects. The system scanned the projects, filtered nearly 10,000 potential issues, and after AI and human review, sent 99 verified reports privately to project maintainers. Maintainers reviewed the reports and fixed 55 vulnerabilities, with 14 of those fixes released to the public. The pilot ran up to November 15, involving both AI and human researchers in the process.

Key details

  • Codename MDASH filtered nearly 10,000 potential issues to 99 verified reports.
  • The pilot involved six open source projects.
  • Maintainers fixed 55 vulnerabilities and released 14 public patches.
  • Seven researchers from Microsoft FORGE Lab manually checked findings.
  • Reports were sent privately to maintainers, each including a suggested fix.

Why it matters

Open source project maintainers now receive fewer, more actionable security reports, reducing the time and effort needed to address vulnerabilities. This approach means maintainers spend less time sorting through false positives and more time reviewing and applying fixes, which can speed up the release of security patches. For organizations relying on open source software, this can lead to faster mitigation of security risks.

Also covered by 1 source


More in Cybersecurity