OpenAI publishes early guidelines on 'safety cases' for frontier AI training runs
OpenAI outlined a push toward structured, evidence-based safety documentation before continuing frontier reinforcement learning training runs, alongside new practices for investigating AI misalignment incidents.