OpenAI has built an autonomous AI pipeline that discovers Linux kernel vulnerabilities, creates working local privilege-escalation (LPE) exploits, and generates candidate patches for human review. In an October experiment the system produced exploits for 948 distinct LPE flaws at a total API cost of $1,213,430, averaging $1,279 per exploit, according to a November 4 research report. The pipeline also generated patches, though early versions sometimes added unnecessary code changes or missed related variants. Human researchers validated root-cause analysis, proof-of-concept exploits and fixes before reporting to maintainers. Of the findings, 61 vulnerabilities were already patched upstream, 43 had staging patches, and 299 were publicly discussed. OpenAI estimates the 948 exploits could be worth $18.9 million to $66.3 million, 16 to 55 times the pipeline’s cost, and plans to extend the work to other critical software through its Patch the Planet initiative. 1