Innovation & Tech

GitHub Pilot Turns 9,791 AI Findings Into 55 Open Source Fixes

GitHub and Microsoft reduced 9,791 possible security issues to 99 reviewed private reports, generating 55 fixes across six projects. Fourteen fixes shipped, but expert review remains unscalable.

By Nathan Brooks

3 min read

GitHub Pilot Turns AI Security Findings Into 55 Open Source Fixes
GitHub Pilot Turns AI Security Findings Into 55 Open Source FixesInnovalabs / pixabay

What's News

  • GitHub’s Microsoft pilot produced 55 fixes across six open source projects as of November 15, 2026.
  • AI triage and human review reduced 9,791 possible issues to 99 private reports; 14 fixes had shipped.
  • Seven Microsoft FORGE Lab researchers manually reviewed every remaining finding.
  • Caddy fixed all 22 reports in a week or less; ImageMagick received 22 and shipped one fix in 15 minutes.
  • Researchers spent about two days per project on manual checks, which GitHub says is not yet scalable.

By Nathan Brooks

GitHub’s security pilot with Microsoft produced 55 fixes across six open source projects as of November 15, 2026. The Codename MDASH trial filtered 9,791 possible issues into 99 private reports, GitHub said on November 17, 2026.

GitHub said the trial turned automated scanning into a manageable private queue rather than a flood of maintainer requests. Fourteen fixes had already shipped in releases. The pilot covered AutoGPT, Caddy, Express, FastMCP, ImageMagick and Netty.

Microsoft’s Codename MDASH system scanned each project with its MAI Cyber model and suggested fixes. A GitHub Security Lab AI triage agent, built on tailored Copilot skills, removed duplicates, checked validity and focused on high-impact bugs.

Each surviving report carried a proof of concept and suggested fix. Seven Microsoft FORGE Lab researchers manually checked every remaining finding before sending it to maintainers through GitHub’s private vulnerability reporting system.

GitHub said maintainers controlled which findings to fix, how to fix them and when to release patches. Reports and patches stayed on GitHub. The company said the arrangement addresses a problem created by cheaper AI scans: duplicate and invalid reports consume time even though remediation still requires human judgment.

GitHub also wants researchers to consult maintainers before pointing another model at a project. It also wants findings submitted with patches.

What Does This Change for Open Source Maintainers?

Maintainers received fewer, better-defined reports while retaining control over remediation. GitHub said some used suggested fixes as written and others adapted them. The first fixes shipped about a week after the initial scan.

GitHub reported that Caddy’s maintainers fixed all 22 reports they received in a week or less. ImageMagick’s two-person maintenance team also received 22 reports. In one case, the maintainers shipped a fix in 15 minutes after a summary pinpointed the relevant lines and showed that an apparently familiar issue was different.

GitHub reported that a report reached AutoGPT maintainers three days after they shipped a broken check and pointed them to the right line. The team went beyond the suggested patch, wrote its version within a day and put it into production less than a week later.

A Netty maintainer put the remaining constraint plainly: “Finding vulnerabilities is no longer our bottleneck. Fixing them is the hard part, and that depends on maintainers.”

Review Capacity Sets the Pace

GitHub called the project sample small and the conclusions early lessons. The company said manual review was not yet scalable: researchers spent about two days per project checking findings by hand. Low-quality reports still reached maintainers every day.

Expansion also needs private-fork CI/CD testing and maintainer access to cyber-capable models, GitHub said. The tests would verify fixes on private forks. Model access would let maintainers test a proof of concept before and after remediation.

GitHub’s recommended standard starts with checking each AI-generated report against a project’s existing knowledge, threat model and security policy. The standard also requires private reporting, a suggested fix and maintainer control over the release pace.

GitHub plans to bring the pilot’s lessons into private vulnerability reporting and security advisories. It will offer AI security scans, with findings checked before private reporting, plus GitHub Copilot Max, to all alumni projects of the GitHub Secure Open Source Fund and extend that support to future Fund sessions.

GitHub and Microsoft are also working to expand researcher review while preserving the manual checks. For maintainers, the practical change is a shorter queue of technically checked findings, with final decisions still theirs. GitHub and Microsoft’s next test is whether expert review can expand without returning the triage burden to maintainers.

Source: unite.ai

Share this article:

More from Nathan Brooks

Nathan Brooks

Show full bio

News editor covering regulation, policy and their impact on entrepreneurship, startups and venture funding at Business Bearings.

2,352 articles

Related articles

« Previous articleNext article »