OpenAI model finds and patches 948 Linux kernel flaws in one run

OpenAI says one of its unreleased cybersecurity models pulled off a full sweep of the Linux kernel: finding, validating and patching 948 privilege escalation flaws in a single autonomous run. The company priced the whole exercise at $1.2 million in API spend. Cointelegraph flagged the claim first.

Privilege escalation bugs are the nasty kind — they let someone with limited access claw their way up to root-level control of a machine. Finding them is hard. Confirming they're real is harder. Fixing them is harder still, and OpenAI says the model handled all three stages without a human holding the wrench.

Divide the bill by the bug count and each fix lands at roughly $1,270 — cheap next to the specialist hours such work usually burns, at least on OpenAI's own math. The model did not come with a name, a release date or any indication of when it might be available.

There's also no word on whether those 948 patches have actually made it into the Linux kernel itself, or who reviewed them. OpenAI's post described the run as autonomous from start to finish, but the company did not detail how the fixes were checked before being counted.

If the numbers hold up, it's a loud signal about where AI-driven security work is heading: not just spotting vulnerabilities at scale, but closing them in the same pass. For now, the only source for the figure is OpenAI, and the model behind it stays behind the curtain.