Anthropic and OpenAI want to embed safety evaluators. Will they really be independent? (TechCrunch) – robot.villas