OpenAI AI pipeline found 948 Linux root exploits for $1.2M

OpenAI says its AI security pipeline built 948 working Linux kernel root exploits in a single run, for about $1.2 million in API credits. That is roughly $1,279 per path to root. OpenAI used the run to find and patch bugs, not to attack anyone. But the lesson for anyone who runs a Linux server is plain: if a low-privilege account can log in to a box that is behind on kernel patches, treat that account as root.

What actually happened

OpenAI detailed the work on Wednesday in a write-up called A thousand fewer paths to root. The Hacker News reported the numbers:

• The model. An "upcoming" cyber-focused model that OpenAI has not released or named.
• The run. In mid-October 2026, the pipeline looked at 2,308 candidate Linux kernel findings and reproduced 1,724 of them within 30 hours.
• The output. 948 local privilege escalation (LPE) exploits that let an unprivileged local user get root on "a commonly used Linux distribution," plus proofs of concept for 291 information-leak bugs.
• The bill. $1,213,430 in API credits, or $1,279 per LPE exploit.
• New techniques. At least 10 Linux exploitation techniques that OpenAI calls novel.
• Humans in the loop. The last stage of the five-stage pipeline is human review. Security researchers check each exploit, its patch and its severity before it goes to kernel maintainers.

OpenAI's conclusion, as quoted: a low-privileged account on a system "that isn't up-to-date on its latest available patches should be considered functionally equivalent to root access for a well-resourced threat actor."

OpenAI did not name the distribution, list CVEs or say which fixes have shipped. We won't guess.

Why Linux root exploits matter for your business

These are local bugs. An attacker needs a foothold first. Small businesses hand out footholds all the time: a shared VPS login for the contractor, an old SSH key for a plugin developer, a web app that gets popped through a vulnerable dependency, a container that shares the host kernel.

The old math was "they got a low-privilege shell, so the damage is limited." At $1,279 per root exploit, that math is gone. Patch speed is now the control that matters.

What we would do this week:

• Turn on automatic kernel security updates (unattended-upgrades on Debian/Ubuntu, dnf-automatic on RHEL-family) and schedule the reboots. A kernel patch does nothing until the box reboots or live-patches.
• List every account that can log in to each server. Remove old contractors, shared logins and stale SSH keys.
• Check the running kernel, not the installed one: uname -r against your distro's latest security release.
• Stop giving agents and CI jobs real shells on production boxes. Give them scoped APIs.

Sources: The Hacker News, OpenAI.