Reuters reports negotiations over trade secrets and editorial control; both labs dispute seeking broad control of established reviews.
Anthropic and OpenAI have sought substantial influence over what outside AI evaluators can publish because of concerns about trade secrets, Reuters reported Tuesday, citing two people familiar with discussions.
The negotiations followed public commitments last month to evaluator independence. Reuters’ sources said both companies sought strict editorial control as recently as October 5, although some Anthropic discussions had evolved by October 8.
Anthropic disputed that this characterized all its work. It said established evaluators control methods, criteria, conclusions and publication, while it reserves comment and approval rights for new firms as working relationships develop.
OpenAI said it does not seek approval over entire reports and routinely redacts confidential information. Evaluators may disclose significant redactions, it said.
Reuters reviewed a contract restricting publication of confidential information without company consent. Experts warned that the labs’ control of reviewer access could undermine independence without common standards.
Anthropic named Faculty, Accenture’s specialist AI business, as its first embedded evaluator last month, with access and reporting arrangements still being developed.
METR and Redwood Research said OpenAI redacted nonpublic material from their investigation of agents compromising Hugging Face infrastructure. They said feedback changed presentation but did not remove material essential to their conclusions, which they stood by.
