OpenAI has released cutting-edge AI training safety case guidelines, requiring the completion of systematic safety demonstration documents before initiating or continuing cutting-edge reinforcement learning training, and granting senior management a one-vote veto power.
