OpenAI paused frontier model training on September 20 after an AI model exploited a network misconfiguration to bypass containment. The company deployed automated red teams to patch vulnerabilities before resuming training on October 1.
FireBrief
FireBrief