OpenAI's autonomous pipeline built 948 Linux kernel root exploits, with patches, in one $1.2M run

On November 4, 2026, OpenAI published A thousand fewer paths to root, describing an autonomous pipeline for vulnerability discovery, exploitation and patching. Scout agents picked 5,034 security-relevant targets, about 2.8 million lines or 7% of the kernel; analyst agents produced 2,308 candidate findings; triage agents reproduced 1,724 of them in isolated virtual machines; and exploit agents built proofs of concept for 291 information leaks and 948 distinct local privilege escalation (LPE) exploits that turn an unprivileged user into root on a stock desktop distribution with standard protections. Human researchers reviewed every result before reporting. OpenAI later confirmed the model was GPT-6 Cyber.

The mid-October run cost $1,213,430 in API credits, or $1,279 per LPE exploit. A July run with GPT-5.6 Sol found 166. On a hardened kernelCTF setup, the pipeline produced 21 LPE exploits for about $32,936, and it also found and exploited a QEMU TCG bug that was already fixed upstream but not in the release build used. OpenAI says its agents used at least 10 exploitation techniques that outside kernel experts had not seen, and it withheld their details.

OpenAI names no CVE IDs or specific fixed vulnerabilities. Following the Linux community's guidance, it began with critical issues and has disclosed 304 critical findings to maintainers with candidate patches, working through Patch the Planet partners for the rest. Within 10 days of the scan, 61 of the exploited bugs were patched upstream, 43 had staging patches and 299 were being reported and discussed publicly. As of November 5, 2026, most of the 948 had no public fix yet.