Anthropic and OpenAI reportedly sought strict editorial control over what outside AI evaluators could publish

Reuters reports that two people said Anthropic and OpenAI sought strict editorial control over outside evaluators’ publications as recently as October 5, despite publicly supporting independent reviews. Anthropic disputes that characterization across all its talks; by October 8, some discussions appeared to have shifted, according to one source. Anthropic says evaluators it has worked with control what they publish but reserves approval rights for new firms. OpenAI says it does not seek approval rights over full report text and redacts confidential information.