Anthropic and OpenAI want to embed safety evaluators. Will they really be independent? September 16, 2026 · TechCrunch Read full story at source