OpenAI says its AI pipeline built 948 Linux kernel root exploits for $1.2M

OpenAI said on Nov. 4 that in one research run with its upcoming cyber model, an autonomous vulnerability discovery, exploitation, and patching pipeline "validated and generated patches for 948 local privilege escalation vulnerabilities in Linux at an API cost of $1.2 million." The write-up describes a prototype in a controlled research environment, not a product or API launch.

The mid-October 2026 run cost $1,213,430 in API credits, or $1,279 per LPE exploit, OpenAI said. Analyst agents produced 2,308 candidate findings within 12 hours. Dynamic validation reproduced 1,724 vulnerabilities within 30 hours. Screening narrowed those to 1,238 LPE candidates and 313 leak candidates. Exploit agents then produced 1,169 working local privilege escalation exploits and 291 information-leak proofs of concept. OpenAI counted 948 distinct patches as the unique LPE total. Scan work and manual analysis finished after 10 days.

Exploits ran in isolated virtual machines with a stock desktop Linux configuration and standard mitigations. Success meant obtaining a root shell. OpenAI does not claim any of the bugs are being used in the wild.

Human researchers reviewed the outputs. OpenAI says it has already sent out hundreds of reports and patches for critical issues and is working with Patch the Planet partners on the rest, alongside Linux maintainers. It does not say those patches have been accepted or merged upstream.

For comparison, OpenAI said a GPT-5.6 Sol run in July found 166 unique LPEs for $513,690 in API credits, and that GPT-5.5 Cyber in June developed 24 LPE exploits. OpenAI claims agents used at least 10 novel Linux exploitation techniques; that claim is the company's own assessment.

Separately, OpenAI noted that while its scan ran, other contributors had already landed 61 upstream LPE fixes, staged 43 more patches, and publicly discussed 299 reports. Those figures are not OpenAI's own patch credits.

OpenAI also floated a hypothetical $18.9 million to $66.3 million bounty valuation by applying historical Pwn2Own and TyphoonPWN Linux LPE payouts to the 948 exploits. That figure is illustrative only, not money paid or awarded.

Sources: OpenAI; OpenAI on X.