OpenAI AI Pipeline Finds 948 Linux Kernel Vulnerabilities
OpenAI used an autonomous AI agent pipeline to identify 948 privilege-escalation vulnerabilities in the Linux kernel and generate corresponding patches for human review.
OpenAI developed an autonomous AI pipeline that identifies software vulnerabilities, creates working exploits for validation, and generates patches for human review. In an October 2026 experiment targeting the Linux kernel, the system identified 2,308 candidate vulnerabilities, resulting in 948 working local privilege-escalation exploits and 291 proof-of-concept demonstrations of information disclosure.
The five-stage process involved vulnerability discovery, dynamic validation, exploitability assessment, patch generation, and human review. The operation cost approximately $1.21 million in API credits, averaging $1,279 per local privilege-escalation exploit. Researchers discovered 10 previously undocumented exploitation techniques, including kernel memory manipulation, page grooming, and data-only attacks.
OpenAI researchers reviewed the root-cause analysis and fixes before coordinating with Linux kernel maintainers to prioritize and apply patches. While some flaws were already patched or publicly discussed, the company has submitted hundreds of reports and patches. OpenAI is withholding specific details on the novel exploitation techniques until they are fully remediated to prevent use by threat actors. The company plans to expand this research to other critical software projects through its Patch the Planet initiative.
