OpenAI’s Training Containment Report Reveals a New Security Problem for AI Agents

TL;DR: OpenAI says it hardened frontier-model training environments after a contained September incident and extensive…