Get the app
Research

OpenAI's AI agents built 948 Linux root exploits for $1.2M

An unreleased OpenAI cyber model found and exploited 948 Linux kernel privilege-escalation bugs in one run, at about $1,279 each.

OpenAI's AI agents built 948 Linux root exploits for $1.2M

OpenAI's Cyber team aimed an autonomous discovery-to-patch pipeline at the Linux kernel in mid-October, running on an upcoming, unnamed cyber model. Agents flagged 2,308 candidate bugs, reproduced 1,724 in about 30 hours, and ended with 948 distinct local privilege escalation exploits, each one a way for an unprivileged user to get root. The run also produced proof-of-concept code for 291 info leaks and turned up at least 10 exploitation techniques that outside kernel experts hadn't seen before.

The bill was $1,213,430 in API credits, or roughly $1,279 per exploit. OpenAI notes that Pwn2Own and TyphoonPWN have paid $20K–$70K for a single Linux kernel root exploit, which would value this haul at $18.9M–$66.3M on paper. The pipeline drafts a patch for every bug, and humans reviewed the findings before anything went to maintainers. Only 61 were fixed upstream by the 10-day mark, though, so most are still open.

OpenAI's warning: for now, a low-privileged account on an unpatched system should be treated as functionally equivalent to root against a well-resourced attacker.

Why it matters: finding kernel bugs now costs about as much as an API bill, and maintainers have to patch faster than attackers can run the same pipeline.

Sources

Independent coverage

Written by an AI pipeline from the sources above. Methodology · Report an error

The day's AI news, in one Telegram post.

The stories that matter for people who build with AI, with the source for every claim. Free, no spam.

Join on Telegram
The daily AI brief, on your phone.

Feed, daily deep-dive and bytes — readable offline, with push alerts for the topics you follow.

Get it on Google Play