Future of work
GitHub and Microsoft send fixes, not more AI bug reports
- Published
- Source
- GitHub Blog
Summary
GitHub and Microsoft ran a pilot to relieve open source maintainers flooded with AI-generated security findings. Microsoft's multi-model scanning harness Codename MDASH scanned six widely used projects, and after AI triage plus hand review by seven Microsoft FORGE Lab researchers, maintainers received only 99 checked private reports with suggested fixes. As of November 15, 55 had been fixed and 14 of those were already in a release, according to the two companies, which also note that human verification does not scale yet, taking researchers about two days per project.
Why it matters for our work
The pilot faces a real shift in maintenance work: AI makes finding issues cheap while fixing still depends on people. Sending only verified reports with patches can cut maintainer workload and offers one model for dividing labor between AI and humans in open source.
Translated from the Korean original. Summaries may be translated and edited. Commentary reflects our perspective; forecasts remain the source’s views.