In a pilot on six open source projects including Express, Caddy and Netty, an AI scanner flagged about 9,791 possible issues. A triage agent and manual checks by Microsoft researchers cut these to 99 private reports with proof of concept and suggested fixes. By 15 November maintainers had fixed 55, and 14 of those fixes were in a release. The point for teams is that filtering and human verification before reporting is what made the output usable.
