Anthropic and OpenAI have sought significant control over what outside evaluators can publish about their AI models, out of concern that findings could reveal trade secrets, Reuters reported on Oct. 13, citing two people familiar with the discussions.
The reported push comes weeks after both companies publicly embraced independent review. Anthropic CEO Dario Amodei said in September that external evaluators should have the right to publish key findings "without editorial control," and OpenAI agreed days later, saying assessors should have "editorial independence," according to Reuters. TechCrunch reported that Amodei's proposal included the right to "publish key findings about risk levels, incidents, practices, and the access they received or didn't receive — without editorial control by Anthropic."
According to Reuters' sources, Anthropic and OpenAI were seeking strict editorial control over evaluators' publications as recently as Oct. 5. By Oct. 8, talks between Anthropic and some evaluators appeared to have evolved, one person told Reuters. The people said the labs worry that evaluator staff embedded inside the companies could take proprietary information to rivals; measures under discussion include having lab employees observe embedded teams and barring evaluators at one lab from working at another at the same time.
Anthropic told Reuters that "seeking significant control" did not accurately characterize all of its discussions with independent evaluators. It said that in current talks with evaluators it previously worked with, it had made clear the outside reviewers control their methods, evaluation criteria, conclusions and what to publish. For new evaluator firms, Anthropic said it reserved the right to comment on and approve reports while the reviewers learn how its models operate.
OpenAI told Reuters it was not seeking approval rights over the full text of evaluators' reports. It said its standard practice is to redact confidential information, and that evaluators can disclose whether major redactions were made. Reuters also reported that existing contract language at AI labs can bar evaluation firms from publishing confidential information without the companies' consent, citing a contract it reviewed.
The evaluators named in the reporting include Faculty, Accenture's specialist AI business, which Anthropic said last month would be its first embedded evaluator, with details on access and reporting still being worked out. "Evaluators are only useful if people can trust what they report, so independence and transparency are critical," Faculty CEO Marc Warner told Reuters. Reuters also cited OpenAI's redactions of a report by the nonprofits METR and Redwood Research on the Hugging Face incident; the groups said they edited structure, emphasis, clarity and tone after OpenAI's feedback but that nothing else important to their conclusions was removed.
Outside the labs, evaluators issued a public letter last month setting minimum conditions, including independent control of findings, broad access and protection from retaliation, Reuters reported, saying it has been signed by more than 200 researchers including Geoffrey Hinton. Rumman Chowdhury, CEO of Humane Intelligence PBC, told Reuters: "Evaluators have this very privileged access, but whether explicitly or implicitly, they know that it comes at a cost."
Independent evaluation was one of four safety measures in the White House Accord on Super Intelligence signed last month, but the accord did not say how reviews should be conducted, Reuters noted. Until standards exist, the terms the labs negotiate now will shape what the public learns about their most powerful models.