Microsoft AI pilot yields 99 verified flaws in six open-source projects
Original title: Microsoft’s AI Uncovers 99 Security Issues in Express, ImageMagick and Other Open-Source Projects
Maintainers had fixed 55 issues by November 15, but only 14 of those fixes were in released software.
GitHub and Microsoft sent 99 verified security reports to Express, ImageMagick, Caddy, AutoGPT, FastMCP and Netty after Codename MDASH, using an earlier MAI Cyber model, flagged thousands of potential flaws. A GitHub Security Lab tool removed duplicates and weak reports, and seven Microsoft FORGE Lab researchers reviewed the rest. Each private report included evidence and a proposed patch. As of November 15, maintainers had fixed 55 issues, with 14 fixes shipped in releases.
Unconfirmed: GitHub describes the pilot as an early test, not proof the process can keep pace across thousands of repositories.
- Caddy fixed all 22 reported issues within a week.
- ImageMagick’s two-person team also received 22 reports and pushed one fix in 15 minutes.
- AutoGPT maintainers wrote their own fix within a day and deployed it in under a week.
- Human review took about two days per project, and the pilot covered only six projects.
- GitHub plans privately delivered reviewed scans for Secure Open Source Fund alumni.
Impact: Unspecified verified security issues, with proposed patches
| Product | Affected |
|---|---|
| OpenJS Foundation Express | — |
| ImageMagick | — |
| Caddy | — |
| AutoGPT | — |
| FastMCP | — |
| Netty | — |
- 1.Enable Dependabot alerts for known vulnerabilities in dependencies.
- 2.Check a public advisory against the installed version before updating.
- Organizations
- GitHub Security LabMicrosoft FORGE Lab
- AI models
- MAI Cyber
From the original article
2 min read at cybersecuritynews.com
Microsoft and GitHub sent 99 verified security reports to six open-source projects, including Express and ImageMagick, after an AI pilot identified thousands of potential vulnerabilities. In its Nov. 17 announcement, GitHub said maintainers had fixed 55 issues as of Nov.
By Abinaya · cybersecuritynews.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.