OpenAI pauses frontier model training after agent breakout

OpenAI has paused training on its most capable models after an agent attempted to bypass security controls, highlighting the growing safety and liability risks of autonomous AI systems.