OpenAI's autonomous kernel vulnerability pipeline

OpenAI's GPT-6 Cyber model, accessed via its Daybreak Red program, powered an autonomous pipeline that found, validated, and generated patches for 948 local privilege-escalation vulnerabilities in the Linux kernel. Scout agents narrowed the kernel's 40 million-plus lines of code to roughly 2.8 million lines for auditing; the system flagged 2,308 suspected bugs, confirmed 1,724 on test machines, and built working exploits for 948. GPT-6 Cyber vastly outperformed its predecessor GPT-5.6 Sol, which found 166 bugs under the same conditions. The model also found 81 previously unknown Firefox vulnerabilities, with Mozilla releasing fixes for many. Total API cost for the kernel run was $1,213,430, with $775,929 attributed to exploit generation; because OpenAI used its own model, that figure may not reflect cash outlay. OpenAI tested the system against a hardened version of Google's kernelCTF, producing 21 LPE exploits over 112 hours at roughly $33,000 in API credits. Human reviewers checked results before patches were submitted to Linux kernel maintainers, and OpenAI is submitting fixes in stages through its 'Patch the Planet' initiative.