GitHub and Microsoft report 55 security fixes from AI scanning pilot
The pilot delivered 99 checked private reports across six open source projects, with proposed fixes; 14 fixes had reached software releases.
Numbers in brackets link to the sources listed below.
GitHub said on 17 November 2026 that its security scanning pilot with Microsoft delivered 99 checked private reports with proposed fixes across six open source projects[1The GitHub BlogMaintainers don’t need more AI bug reports. They need fixes.Open sourceSources [1]]. SecurityBrief reported that maintainers had fixed 55 issues and released 14 fixes[2SecurityBrief USGitHub & Microsoft pilot AI bug scans in open sourceOpen sourceSources [2]].
According to SecurityBrief, the system initially flagged 9,791 potential issues, which underwent automated filtering and human review before reports reached maintainers[2SecurityBrief USGitHub & Microsoft pilot AI bug scans in open sourceOpen sourceSources [2]]. The pilot paired Microsoft’s Codename MDASH scanning system with GitHub Security Lab’s AI screening and manual checks by Microsoft researchers[2SecurityBrief USGitHub & Microsoft pilot AI bug scans in open sourceOpen sourceSources [2]]. It covered AutoGPT, Caddy, Express, FastMCP, ImageMagick and Netty[2SecurityBrief USGitHub & Microsoft pilot AI bug scans in open sourceOpen sourceSources [2]].
Microsoft describes MDASH as a code scanner within Microsoft Defender whose specialised AI agents work through stages to identify vulnerabilities and establish evidence for them[3Microsoft LearnCodename MDASH overview (preview)Open sourceSources [3]]. In its preparation stage, the system prioritises files by assessing risk through code complexity and relationships between function calls[3Microsoft LearnCodename MDASH overview (preview)Open sourceSources [3]].
GitHub said maintainers already face a flood of AI-generated security submissions, including duplicate reports and findings that do not represent vulnerabilities[1The GitHub BlogMaintainers don’t need more AI bug reports. They need fixes.Open sourceSources [1]]. While AI makes identifying potential flaws faster and cheaper, fixes still require human judgment, it said[1The GitHub BlogMaintainers don’t need more AI bug reports. They need fixes.Open sourceSources [1]]. GitHub urged anyone scanning a project with a new model to seek maintainers’ agreement first and provide fixes alongside findings[1The GitHub BlogMaintainers don’t need more AI bug reports. They need fixes.Open sourceSources [1]].
Sources
- 1Maintainers don’t need more AI bug reports. They need fixes.
PrimaryThe GitHub BlogAccessed
- 2GitHub & Microsoft pilot AI bug scans in open source
IndependentSecurityBrief USAccessed
- 3Codename MDASH overview (preview)
PrimaryMicrosoft LearnAccessed
ITANU’s newsroom system produced this article with AI and checked it against the sources listed above. How we work


