AI has made the detection of possible vulnerabilities faster and cheaper. However, fixing them still requires human judgment, and in open source software, this task falls to maintainers. Many are already receiving a deluge of AI-generated security reports, including duplicates and reports that do not represent vulnerabilities.

Those responsible for maintenance are under intense pressure due to this avalanche of reports. As Christian Grobmeier, one of the maintainers of Log4j, has said, the AI ​​that defends open source has to be better than the AI ​​that attacks it.

AI should not increase the workload of maintainers. Therefore, GitHub and Microsoft have carried out a pilot project: could AI help them implement fixes instead of giving them more problems?

GitHub and Microsoft are using Codename MDASH, Microsoft’s multimodal scanning tool, to find and fix security vulnerabilities in widely used open source software, thereby strengthening the foundation on which much of the world’s code is built.

Find and fix vulnerabilities in software

The pilot program had two goals: send only relevant findings and never send an unverified finding to a maintainer. This is how it behaved:

  1. Find. The artificial intelligence system codenamed MDASH, developed by Microsoft Security, analyzed each project using Microsoft’s MAI Cyber ​​model and suggested solutions for the problems detected.
  1. Filter. An AI triage agent from GitHub Security Lab, based on Copilot’s custom capabilities, removed duplicates, checked validity, and focused on high-impact bugs, so maintainers never had to review the raw scan results.
  1. Verification. Researchers at the Microsoft FORGE lab manually verified all remaining findings.
  1. Report privately. The reports reached maintainers through GitHub’s private vulnerability reporting system, which included a proof of concept and a suggested fix.
  1. Correction. Those responsible for maintenance decided what to fix, how to fix it, and when.

The reports and corrections remained on GitHub, where those responsible for maintenance already work.

As a result, the Codename MDASH project, which uses Microsoft’s MAI Cyber ​​model, identified thousands of potential issues across the six projects. After AI classification and manual review by seven Microsoft FORGE Lab researchers, maintainers received 99 private reports. As of November 15, they had fixed 55 of them, and 14 were already included in a release.

Help improve supply chain security

Through Codename MDASH, six popular open source projects have been analyzed, converting 9,791 initial findings into 99 verified fixes, which were communicated to project managers through private reports. Maintainers are already implementing these fixes, helping to improve the security of the software supply chain.

This initiative adds to the support that GitHub has been offering for years to people who maintain open source projects. The six projects analyzed are part of the GitHub Secure Open Source Fund, a program that provides each project with $10,000 through GitHub Sponsors, in addition to practical training in cybersecurity.

Now, GitHub will extend these AI-based security analyzes to all projects that have participated in the program, which already has 216 participants. Any possible problems detected will be reviewed and verified before being communicated privately to those responsible.