OpenAI Reports 948 Linux Kernel Flaws That Could Enable Local Root Access

OpenAI says an AI-assisted research pipeline identified and validated 948 unique Linux kernel vulnerabilities that could allow low-privileged local users to gain root access. The research disclosure describes a mid-October run combining attack-surface discovery, candidate analysis, isolated reproduction, and exploitation testing against common kernel protections, with human researchers reviewing results before reporting. Secondary reporting puts the run’s cost at approximately $1.21 million and says testing also uncovered previously unseen exploitation techniques and a QEMU vulnerability in the testing environment.

OpenAI reportedly generated patches for all 948 findings, but those proposed fixes were not equivalent to deployed protection: within the reported 10-day period, 61 vulnerabilities had been patched upstream, 43 had staged patches, and 299 were under public discussion. The findings underscore the potential scale of AI-assisted vulnerability discovery while leaving substantial validation and remediation work for maintainers. Linux operators should prioritize applicable kernel updates, reboot into patched kernels, and restrict unnecessary unprivileged attack surfaces, including user namespaces where operationally feasible.