OpenAI pauses frontier RL training to tighten AI safety

OpenAI paused reinforcement learning training for two weeks and halted its largest planned frontier RL run after internal evaluations showed its upcoming Astra model may reach 'critical' cybersecurity capability. New safeguards include stronger sandboxes, network isolation, and a monitoring system that pages teams within 30 minutes, consuming ~20% of inference compute.

Featured · Amelia Glaese