Overview
- OpenAI disclosed Wednesday that a mid‑October autonomous run against the Linux kernel reproduced 1,724 vulnerabilities, generated 1,169 working local privilege escalation (LPE) exploits and mapped those to 948 distinct LPEs at a reported API cost of $1,213,430.
- The system used scout agents to narrow the audit to 5,034 security‑relevant targets (about 2.8 million lines, roughly 7% of the kernel) before analyst and triage agents produced candidate findings, validated bugs in sandboxes, and produced exploits and candidate patches.
- OpenAI says human researchers reviewed results before disclosure and that it has sent hundreds of reports and patches to maintainers with 61 fixes accepted upstream and 43 more in staged processes at the time of announcement.
- The researchers sampled exploits and identified at least 10 exploitation techniques that third‑party kernel experts said were previously undocumented, raising concerns that new tactics will expand the attack surface for systems that are not promptly patched.
- Because the pipeline cut the per‑exploit cost to about $1,279 compared with historical bounty valuations, defenders now face a practical challenge of scaling human review and deployment of fixes to prevent rapid weaponization on billions of Linux‑based devices.