Anthropic and OpenAI reportedly sought significant control over what outside AI safety evaluators can publish

Anthropic and OpenAI want outside safety evaluators inside their labs, but Reuters reports both sought significant control over what those evaluators could publish as recently as October 5. Talks with some Anthropic evaluators appeared to evolve by October 8. Anthropic says evaluators it previously worked with decide what to publish but reserves approval rights for new firms. OpenAI denies seeking full-report approval and says it redacts confidential information.