Reuters: Anthropic and OpenAI seek control over outside AI evaluators' publications

Anthropic and OpenAI have pushed for a say over what outside reviewers of their AI models can publish, Reuters reported, citing sources, over fears of exposing trade secrets. Ideas under discussion include lab staff overseeing evaluator teams and banning reviewers from working at two labs at once. Both labs had recently promised reviewers a free hand.

• Anthropic and OpenAI fear evaluator reports could expose trade secrets
• Talks include lab oversight of evaluator teams and a ban on working at two labs
• Until at least October 5 both labs wanted strict control over publications
• OpenAI already cut details from a METR and Redwood Research report on Hugging Face hacking