AI for Business

GitHub and Microsoft Pilot Turns AI Scans into 99 Checked Fixes for Open Source Maintainers

Open source maintainers are drowning in AI-generated security findings, many of them duplicates or not vulnerabilities at all. A pilot from GitHub and Microsoft tested a different approach: what...

Share:

Open source maintainers are drowning in AI-generated security findings, many of them duplicates or not vulnerabilities at all. A pilot from GitHub and Microsoft tested a different approach: what if AI delivered fixes instead of more reports?

The pilot used Codename MDASH, Microsoft's multi-model agentic scanning harness built on its MAI Cyber model, to scan six widely used open source projects. An AI triage agent from the GitHub Security Lab then removed duplicates and checked validity, and seven Microsoft FORGE Lab researchers reviewed findings by hand. Maintainers received 99 private reports through GitHub's private vulnerability reporting, each with a proof of concept and a suggested fix.

As of November 15, maintainers had fixed 55 of the reports, with 14 fixes already in a release. Caddy's maintainers cleared all 22 of their reports in a week or less, and an ImageMagick team of two pushed a fix in 15 minutes after a report pointed to the exact lines of code.

Maintainers quoted in the announcement said the reports cut through the noise and surfaced obscure bugs earlier scans missed. The pilot's organizers acknowledged the limits: six projects is a small sample, and seven researchers spending about two days per project on manual checks does not scale yet. They urged anyone running AI scans against open source to check findings first, report privately, and include a suggested fix.

Source: GitHub News

Ready to Modernize Your Business?

Get your AI automation roadmap in minutes, not months.

Analyze Your Workflows →