IT Brief New Zealand - Technology news for CIOs & IT decision-makers
New Zealand
GitHub & Microsoft pilot AI bug scans in open source

GitHub & Microsoft pilot AI bug scans in open source

Wed, 18th Nov 2026 (Today)
Mara Sugue
MARA SUGUE News Editor

GitHub and Microsoft have disclosed early results from a pilot that uses artificial intelligence to identify and help fix security flaws in open source software. Across six widely used projects, the system produced 99 verified private reports.

The pilot combined Codename MDASH, Microsoft's multi-model scanning system, with an AI triage process from GitHub Security Lab and manual checks by Microsoft researchers. The goal was to cut the number of raw alerts reaching maintainers and instead send private reports with proof-of-concept material and suggested fixes.

The system flagged 9,791 possible issues across the six projects. After automated filtering and manual review, maintainers received 99 private reports. More than half had already been fixed when the figures were published.

Of the 99 reported issues, 55 had been fixed and 14 were already included in a software release. The six projects in the pilot were AutoGPT, Caddy, Express, FastMCP, ImageMagick and Netty.

The work addresses a growing problem for open source projects, whose maintainers are increasingly dealing with large numbers of AI-generated vulnerability submissions, including duplicates and issues that are not actually security flaws. Many of those maintainers are volunteers or work in small teams.

Under the process described by the companies, Codename MDASH first scanned each code base using Microsoft's MAI Cyber model and proposed fixes. A triage agent built by GitHub Security Lab then removed duplicate findings, tested their validity and narrowed the list to higher-impact issues. Researchers from Microsoft FORGE Lab manually checked the remaining reports.

Only then were reports sent to maintainers through GitHub's private vulnerability reporting system. Each included a suggested fix, while maintainers retained control over whether and when to patch the issue.

Maintainer burden

The pilot was designed around two principles: send only findings that matter, and do not send unverified reports to maintainers. That reflects wider concern across the software industry that AI has sharply lowered the cost of finding potential flaws, but not the cost of validating and fixing them.

The first fixes were shipped roughly a week after the initial scans. In one example highlighted by the companies, Caddy maintainers fixed all 22 reports they received within a week or less.

On ImageMagick, which was described as being maintained by a two-person team, the pilot generated 22 reports. One maintainer said the format of the reports reduced the time needed to assess and fix some issues.

"Normally, a security advisory would take us around 30 minutes to 1 hour on average to read, understand, and make a fix. It is a huge difference. Codename MDASH was able to find unknown, obscure security bugs in our project with succinct reports, and it filled in gaps from some prior reports that hadn't been noticed," said Grobmeier.

Another maintainer assessment cited by the companies came from FastMCP. "The Codename MDASH scan provided unique insights into the security of our project above and beyond what AI scanners have provided us in the past," said Grobmeier.

A further maintainer comment focused on the volume of low-quality reports. "Codename MDASH is very useful for cutting through the noise of vulnerability reports," said Grobmeier.

Scaling up

The pilot remains limited in scale and still depends heavily on human review. Seven Microsoft FORGE Lab researchers spent about two days per project checking findings by hand, a level of scrutiny that may be difficult to expand broadly without additional resources.

That challenge is one of the project's main lessons. While AI can generate large numbers of possible findings cheaply, open source security still relies on careful human validation and maintainers' own judgment when deciding how to patch code.

GitHub is now extending AI security scans to all alumni projects in its Secure Open Source Fund, with findings reviewed before they are reported privately. The fund has supported 216 projects so far, and the participating repositories account for software downloaded more than 20 billion times a month.

Each project in the fund receives USD $10,000 through GitHub Sponsors, alongside security training. More than 300 maintainers in the programme have fixed more than 4,000 vulnerabilities. Broader data also shows more than 850 secret leaks prevented or resolved and more than 1,500 vulnerable dependencies updated.

The six projects in the MDASH pilot were all alumni of that fund. GitHub said the support package for alumni will now also include AI security scans and access to Copilot Max.

The broader significance lies in how flaws in open source software can affect commercial software supply chains far beyond the original projects. When maintainers publish a security advisory and it is reviewed for the GitHub Advisory Database, Dependabot can alert downstream repositories that rely on the affected package.

"Maintainers do the fixing. Our job is to make sure they never do it alone," said Wolf.