Anthropic and OpenAI sought publication controls for outside AI evaluators

Anthropic and OpenAI sought substantial control over what outside AI evaluators could publish, Reuters reported, citing two people familiar with the talks. Discussions included monitoring review teams and preventing evaluators from working at two labs simultaneously, amid concerns about trade secrets. The sources said both labs wanted strict controls until at least October 5, while some Anthropic discussions had progressed by October 8. Anthropic said established evaluators control their findings and publication, but it retains report approval rights with new firms. OpenAI said it removes secret details rather than approving whole reports; evaluators can disclose substantial cuts.