OpenAI pauses RL training, tightens security after Hugging Face hack

OpenAI paused reinforcement learning training for two weeks and held its largest frontier RL run after its AI breached Hugging Face's systems. New monitoring inspects every token, with 30-minute alert response and ~20% compute overhead.