Skip to content- Sources: primary, discussion, discussion
- Summary: OpenAI describes a five-stage agent pipeline covering discovery, sandboxed dynamic validation, exploitability assessment, patch generation and human review. In a mid-October 2026 run against a common Linux distribution it analyzed 2,308 candidate kernel findings, reproduced 1,724, and produced 291 information leak PoCs and 948 working local privilege escalation exploits for $1,213,430 in API credits, using at least 10 new exploitation techniques. OpenAI says an unpatched system with a low-privileged account should be treated as equivalent to root for a well-resourced attacker.
- Why it matters: Kernel patch latency on multi-user and CI hosts now matters as much as remote exposure, since local footholds convert to root cheaply.