OpenAI says autonomous security agents found 948 distinct Linux kernel vulnerabilities and created working exploits for the flaws. The October experiment consumed $1,213,430 in API credits, about $1,279 per vulnerability. The vulnerabilities could let attackers with limited access to a vulnerable machine gain root privileges and full control. The agents also drafted patches, but human reviewers and software maintainers remain responsible for checking and implementing fixes. OpenAI warned that AI could increasingly help attackers escalate privileges on outdated operating systems.