OpenAI conducted an automated security audit of the Linux kernel, a project that cost the company $1.21 million. The initiative utilized a five-stage pipeline to scan over 40 million lines of code, identifying 948 unique vulnerabilities related to privilege escalation.
The automated process began with a 12-hour scan that flagged 2300 potential issues. Subsequent analysis by AI agents over 30 hours confirmed 1724 vulnerabilities, with a 3 percent false positive rate. The system generated 1169 privilege escalation exploits and 291 information disclosure proofs.
Beyond identifying existing flaws, the AI agents discovered 10 previously undocumented exploitation methods, including data-only attacks and kernel memory manipulation. OpenAI has withheld technical details on these specific vulnerabilities until they are patched.
The research team verified all exploits and patches before coordinating with the Linux kernel community. While some vulnerabilities were already known or partially addressed by developers, the automated system significantly accelerated the identification and remediation process.
