HAProxy Lead Developer on AI Tools for Finding Bugs in Open Source Software
On The Use of AI-Based Tools to Find Bugs in Open Source Software
HAProxy lead developer and Linux Kernel security team member reports receiving hundreds of AI-assisted security reports that lack reproducers and patches, and describes limited access to Anthropic's strongest models despite completing its Cyber Verification Program. He says two batches of Mythos-generated HAProxy reports, 30 to 50 entries each, were unusable without commit IDs or code excerpts, and that Opus 4.8 was used to decode them, with about one-third of proposed fixes merged after rework. He argues safeguards on Fable 5 and later models block legitimate defensive work, that higher access tiers favor exploit sellers over maintainers, and that a low-cost subscription for public-project work would help; HAProxy currently has 300 to 400 pending AI-generated reports.
Source:Hacker News · AI · haproxy.com