The idea that Anthropic, OpenAI and other AI companies could “embed” researchers from outside safety groups into their ...