948 root exploits for Linux, at $1,279 each

OpenAI has described an automated pipeline that finds, validates, exploits and patches vulnerabilities, and the numbers from its Linux test are the point. In a run in mid-October, using an upcoming cyber model, the pipeline identified 2,308 candidate findings in the Linux kernel, dynamically reproduced 1,724 vulnerabilities, and produced 948 working local privilege escalation exploits, each letting an unprivileged local user gain root, plus proofs of concept for 291 information leaks. The targets were stock desktop installations of a commonly used distribution, not weakened systems.

The run cost $1,213,430 in API credits, or $1,279 per working exploit. OpenAI sets that against historical competition bounties of $20,000 and $70,000 for a Linux kernel privilege escalation, which would value the 948 exploits at $18.9 million to $66.3 million. On kernelCTF, a hardened kernel with reduced attack surface, a separate run costing about $32,936 produced 21 privilege escalation exploits; for comparison, 26 confirmed zero-day exploits have been accepted by kernelCTF in all of 2026. OpenAI says its agents discovered and used at least 10 novel Linux exploitation techniques.

Each finding comes with a root-cause analysis, a minimal crashing proof of concept, a candidate patch and a full exploit, and human researchers review the results before anything is reported. Following the kernel community's guidance, OpenAI began with the critical vulnerabilities and is working with partners in its Patch the Planet initiative on further reports. While the scan ran, it observed more than 100 Linux privilege escalations patched or staged by other contributors within 10 days, with around 300 more under public discussion.

The conclusion it draws is blunt. For the near and medium term, OpenAI writes, a low-privileged account on an operating system that is not fully patched should be treated as functionally equivalent to root access for a well-resourced attacker. It adds that the newer model found more than five times as many novel privilege escalations as a run with GPT-5.6 Sol in July, despite three months of hardening in between.