OpenAI reproduces 1,724 Linux vulnerabilities… confirms privilege escalation in 948 cases

OpenAI reproduced 1,724 Linux kernel vulnerabilities in a test environment and confirmed 948 distinct local privilege-escalation flaws, in which a user with low-level access can obtain root permissions. The automated security study, run in mid-October against the then-latest Linux kernel, narrowed roughly 2.8 million lines of security-related code—about 7% of the total kernel—and surfaced 2,308 vulnerability candidates. The pipeline spanned five stages: vulnerability discovery, dynamic verification, exploitability assessment, patch generation, and researcher review, with candidates validated in isolated virtual machines. The result matters because it shows AI-driven auditing can triage kernel code at a scale manual review cannot match, and the next thing to watch is how many of the generated patches are accepted upstream and whether the approach generalizes beyond Linux.