Anthropic and OpenAI want to embed safety evaluators. Will they really be independent?

Anthropic and OpenAI want to embed independent safety evaluators inside their AI labs. Researchers welcome the unprecedented access, but warn meaningful oversight requires...