An autonomous pipeline powered by an unreleased OpenAI cyber model built working root exploits for 948 unique Linux kernel vulnerabilities in a single run that cost $1,213,430 in API credits, SecurityWeek reported on November 51.
The mid-October scan targeted the most recent code in the kernel's master branch. AI agents first narrowed the codebase to security-relevant areas reachable by low-privileged users, flagged 2,308 potential bugs, and confirmed 1,724 of them. Exploits were tested on virtual machines running an unmodified desktop Linux setup with common protections — KASLR, SMEP, and SMAP — turned on.
The agents produced 1,169 working local privilege escalation (LPE) exploits and 291 proofs of concept for information leaks. OpenAI said the 948 figure reflects the number of separate patches created for the bugs behind those exploits. Before anything went to kernel maintainers, OpenAI's security researchers checked the exploits, fixes, and proposed severity ratings.
OpenAI said the agents discovered and used at least 10 novel Linux exploitation techniques, according to the company's own assessment. Kernel exploitation experts outside OpenAI were not familiar with them either, and the full technical details are being kept private for safety reasons.
Spread across the 948 exploits, the total bill comes to roughly $1,279 apiece, according to SecurityWeek's reporting. An earlier scan in late July, powered by GPT-5.6 Sol, yielded 166 unique LPEs at a cost of $513,690.
ANALYSIS The October run surfaced 948 unique LPEs compared with 166 in July while total cost rose from $513,690 to $1,213,430, indicating a substantial gain in exploit yield per dollar with the newer, unreleased model.
OpenAI recently closed roughly $11 billion in new funding at a valuation of $1.25 trillion[2].
ANALYSIS The $1.2 million price tag for a single automated security sweep is modest relative to that capital base, but the demonstrated capability to surface nearly a thousand kernel-level exploits autonomously carries direct implications for both defensive security tooling and the dual-use risk calculus around frontier cyber models.