OpenAI AI agent finds 1,724 Linux kernel vulnerabilities, generates 948 exploits

Cointime update: OpenAI’s automated vulnerability research pipeline identified 2,308 candidate Linux kernel flaws and confirmed 1,724 through dynamic reproduction, generating 948 local privilege escalation (LPE) exploits that let unprivileged users gain root access. The second full scan cost about $1.21 million in API credits. OpenAI is submitting the findings and patches to Linux kernel maintainers, and third-party kernel experts confirmed the agent used 10 previously unseen exploitation techniques.