OpenAI says it demoed an autonomous pipeline that finds, reproduces and exploits local privilege escalation flaws in the Linux kernel, the core running Android, cloud and servers, where an unprivileged local user can gain root.
I think cheap, systematic root finding just ended the era of scarce local 0-days. OpenAI says a single mid-October run with an upcoming cyber model dynamically reproduced 1,724 vulnerabilities from 2,308 candidates and produced 948 working local privilege escalation exploits on a stock desktop distribution. That is not a lab trick.
The pipeline's efficiency is the first hard evidence. Scout agents narrowed the 40-million-line kernel to 5,034 audit targets, about 2.8 million lines or 7 percent of the codebase, for $3,426 in API credits, OpenAI said. Analyst agents then scanned those targets in 12 hours for $316,549 and triage agents reproduced 74.7 percent of candidates in 30 hours for $73,330, with only 3.1 percent failing as incorrect analysis.
The second is exploit economics. The full run cost $1,213,430 in API credits, or $1,279 per LPE, according to OpenAI, and AlphaSignal described the result as 948 unique privilege escalation bugs found in a single run at that average price. OpenAI said the historical Pwn2Own and TyphoonPWN bounties of $20,000 and $70,000 per Linux LPE would value those 948 exploits at $18.9M to $66.3M, or 16x to 55x the pipeline cost. On a hardened kernelCTF target with reduced attack surface, it still got 21 LPEs for about $32,936, against 26 confirmed 0-day LPEs accepted by kernelCTF in 2026.
The fair counter is that local is not remote, and defenders are not standing still. OpenAI notes these exploits need a low-privileged local account and modern mitigations like SMEP, SMAP and KASLR were enabled, and within 10 days of scanning over 100 Linux LPEs were patched or staged with about 300 more under discussion. Candidate patches also need human review before maintainers accept them.
The daily brief
Get the free daily brief with the day's AI news, new AI startups and funding rounds every morning.
Join 64,000 other subscribers.
That concession does not save scarcity. OpenAI said the newer cyber model found more than five times as many novel LPEs as a GPT-5.6 Sol run in July despite three months of hardening, and agents used at least 10 novel exploitation techniques, according to the company.
If you ship Linux and patch monthly, assume a well resourced attacker with any local foothold already has root.
