Trail of Bits' AI patches introduced 21 regressions across 130 merged PRs

AI agent fixed the bug, but what else did it change?

Trail of Bits reviewed every Patch the Planet patch its engineers authored and maintainers merged upstream and found 21 regressions across more than 130 PRs — 16 functional, one performance, four build/test, none breaking a security protection. The firm is releasing two plugins: patching-bugs, which reproduces bugs, writes fixes, and proves them with a three-arm test, and regression-hunt, which traces diffs to real interfaces and replays proof-of-concepts on live deployments. In preliminary tests, regression-hunt cut false positives 25% versus a baseline Codex review prompt.