Exclusive-Behind AI industry’s call for independent reviews, a fight over what gets made public

Anthropic and OpenAI are seeking control over what outside evaluators can publish, citing concerns over trade secrets, according to two sources familiar with the discussions. Anthropic CEO Dario Amodei has stated that external evaluators should have the right to publish key findings without editorial control, while OpenAI agreed to the same in October.

The companies worry that evaluators working inside their facilities might leak proprietary information to competitors. Possible measures to prevent leaks include having lab employees observe embedded teams and preventing evaluators from working simultaneously at different labs. However, what labs consider crucial intellectual property can overlap with information relevant to assessing a model's risk, leading to arguments over what should be redacted.

Experts emphasize the importance of evaluator independence as Anthropic and OpenAI choose who conducts the reviews. The negotiations between the AI companies and the evaluators could determine how the world scrutinizes this potentially dangerous technology.