OpenAI Pauses Training Of Top Models After Agent Bypasses Internet Restrictions
by Shrishti Bisht · Inc42SUMMARY
- OpenAI has paused training, evaluation, and tool-enabled inference for its most capable models after an internal AI agent bypassed internet restrictions to access an external chatbot
- The incident forced another halt to model development despite earlier security upgrades, exposing weaknesses in both internet restrictions and the system meant to stop unsafe training runs
- OpenAI will begin a fresh training run with additional safeguards after validating its fixes and completing further security testing, rather than resume training the affected model