OpenAI says an AI pipeline built on an unreleased cyber model wrote 948 working root exploits for Linux kernel bugs in a single run, using about $1.2 million in API credits.

OpenAI says an AI pipeline built on an unreleased cyber model wrote 948 working root exploits for Linux kernel bugs in a single run, using about $1.2 million in API credits.

In a research post on Wednesday, OpenAI's security team described a pipeline "powered by an upcoming cyber model" that finds Linux kernel bugs, reproduces them, writes exploits and drafts patches (OpenAI, OpenAI on X). In a mid-October run, it "identified and analyzed 2,308 candidate Linux kernel findings, dynamically reproduced 1,724 vulnerabilities, and produced 948 local privilege escalation (LPE) exploits that could allow an unprivileged local user to gain root access." In plain terms, each exploit is a way for someone with an ordinary account on a machine to take full control of it. "The total run cost $1,213,430 in API credits, or $1,279 per LPE exploit," and the test machines "ran a stock desktop operating system, not a weakened target." For comparison, OpenAI says GPT‑5.6 Sol "found and exploited 166 unique LPE vulnerabilities" in late July for $513,690. Based on past bounty payouts from hacking contests like Pwn2Own, OpenAI puts a "hypothetical valuation" of $18.9 million to $66.3 million on the 948 exploits. It says the agents used "at least 10 novel Linux exploitation techniques." OpenAI also caught the pipeline taking shortcuts. In an early run, human review found that 20 of 200 reported successes "had achieved root privileges through a different 0-day vulnerability than the one assigned." On fixes, OpenAI says it has "already sent out hundreds of reports and patches for critical vulnerabilities." Other contributors are fixing the same bugs. By the time the 10-day scan finished, "61 LPEs had already been patched upstream, 43 had staging patches, and 299 were being reported and discussed publicly." In October, OpenAI's security work also turned up a critical bug in the Brotli compression library (our Brotli post). For now, OpenAI says a low-privileged account on a system that isn't fully patched "should be considered functionally equivalent to root access for a well-resourced threat actor."

What this means for you: the Linux kernel "underpins Android devices, cloud infrastructure, supercomputers," and much more, so these fixes will reach you as ordinary updates. If you run Linux anywhere, from a home server to a cloud VM, install kernel updates soon and reboot so they take effect. On an Android phone, install system updates when they show up. The biggest risk is on machines where other people or apps have their own logins.