OpenAI says an upcoming cyber model found 948 distinct Linux kernel root exploits for $1.2 million
An autonomous pipeline produced 1,169 working privilege-escalation exploits in a mid-October run, at about $1,279 each; OpenAI says it is now patching them upstream.
Why it matters
OpenAI ran an autonomous vulnerability pipeline on its unreleased cyber model against the Linux kernel. Of 2,308 candidate findings, 1,724 reproduced, yielding 1,169 working local privilege-escalation exploits that map to 948 distinct bugs, for $1.21 million in API credits. In July, GPT-5.6 Sol found 166 for $513,690. Only 61 are patched upstream so far, with 43 staged and 299 under discussion. Watch how fast the remaining fixes land and when the model ships.
Key facts
- Mid-October 2026 run with an upcoming OpenAI cyber model
- 2,308 candidate findings, 1,724 reproduced
- 1,169 working privilege-escalation exploits covering 948 distinct bugs
- $1,213,430 in API credits, about $1,279 per exploit
- 61 patched upstream, 43 staged, 299 under discussion
- July comparison: GPT-5.6 Sol found 166 for $513,690