OpenAI details a Linux kernel security research pipeline that found 948 local privilege escalation exploits

OpenAI published a Nov. 4 article describing a controlled research pipeline for Linux kernel security work. The system scanned 5,034 security-relevant audit targets, covering about 2.8 million lines of code, or roughly 7% of the kernel, then produced 2,308 candidate findings and reproduced 1,724 in isolated environments. In a mid-October 2026 run, OpenAI says it generated 291 information-leak proofs of concept and 948 working local privilege escalation exploits, with total API spend of $1,213,430. It also later developed 948 distinct patches for maintainer review.