OpenAI's AI finds 948 Linux root exploits

🔥 What happened OpenAI ran an upcoming cyber model against the Linux kernel and produced 2,308 candidate findings, dynamically reproduced 1,724 vulnerabilities, and generated 948 working local privilege escalation exploits that grant root. Human researchers reviewed the outputs and reported flaws with candidate patches to kernel maintainers. 💡 Why it matters Scout agents flagged just 7% of the 40-million-line kernel (2.8M lines) as worth auditing — discovery cost $3,426, analysis $316,549. The bottleneck has shifted from finding bugs to triaging and validating them: no team can manually process 2,308 findings. ⚡ Our take This isn't benchmark theater — it's a warning shot. When a model mass-produces hundreds of working root exploits, patch validation becomes the new security discipline for attackers and defenders alike.