Why it matters
Trail of Bits describes supervising GPT-5.5-Cyber as it built ASan and UBSan variants, derived seed corpora, and wrote fuzz harnesses for roughly a dozen zlib entry points in one day. The useful result is the workflow and its emphasis on reachability and reportability; vulnerability details remain under coordinated disclosure and the speed comparison is the authors' estimate.
My takeaway: For AI-assisted fuzzing, define the vulnerability class and reportability criteria up front, require sanitizer-backed reproduction, test compile-time variants and realistic state transitions, and keep a human maintainer in the disclosure loop. Measure confirmed, reachable findings—not crashes or model-generated severity labels.