GitHub and Microsoft have reported the results of an AI-assisted security audit covering six open-source projects, including Express, Caddy, and ImageMagick.
The process involved automated filtering of potential issues followed by manual verification, resulting in 99 confirmed reports submitted privately.
This report details the collaborative workflow, where maintainers reviewed and patched a majority of the findings, with 14 fixes already in production releases.
According to a GitHub announcement on November 17, 2026, maintainers had fixed 55 of the 99 issues as of November 15. However, only 14 of those fixes had been included in released versions. In other words, fixing a problem isn't enough on its own; it also needs to be released in a version that users can use.
The six programs were Express, ImageMagick, Caddy, AutoGPT, FastMCP, and Netty. Their maintainers had approved the audit before it began, as part of the GitHub Secure Open Source Fund.
Microsoft used MDASH, a previous version of the MAI Cyber model, to examine the programs and suggest fixes. A tool from GitHub Security Lab removed duplicate and unconfirmed reports, and prioritized issues that could have the greatest impact. Seven researchers from Microsoft FORGE Lab then manually reviewed the remaining findings. Microsoft had previously used MDASH to identify security issues in Windows.
Reports were submitted privately through GitHub's security reporting system. They contained details about each issue and a suggested fix. Maintainers could accept it, change it, or write their own, and decide when to release it.
↳
READ ALSO
HACKS.GR / RELATED
Vulnerabilities & Security Updates
Computer security flaw allows PIN to be broken with physical access
↗
Caddy fixed all 22 reported issues within a week. The two-person team that maintains ImageMagick also received 22 reports. In one case, the report explained how the new issue differed from a similar issue the team had previously encountered and pointed to the relevant point in the program. The team fixed it within 15 minutes. The finding was separate from previous ImageMagick issues involving computer memory.
The AutoGPT team received a report about a new security check three days after its release. They wrote their own fix within a day and had it up and running in less than a week. Express maintainers said the reports helped them separate the meaningful findings from the large volume of other reports.
The test covered only six programs, and human review took about two days for each. This is a first test, not proof that the same process can be applied at the same rate to thousands of programs.
GitHub plans to offer AI-powered, human-reviewed audits to anyone who has completed the Secure Open Source Fund. The findings will be delivered privately.
GitHub recommends that developers who rely on open source programs enable Dependabot notifications for known security issues in the programs and tools their project uses. When a public security announcement is reviewed and added to the GitHub database, Dependabot can notify projects that use the affected program. This way, teams can check whether their version is affected and whether the next update includes the fix.
>Stay connected with Hacks.gr
Set us as a preferred source on Google and find us on social media.
Preferred source
↗
end of article
Top comments (0)