This is a Reuters exclusive based on two people familiar with the discussions, who were not named. Anthropic and OpenAI both dispute parts of it; their responses are below.
Anthropic and OpenAI have sought significant control over what outside evaluators can publish about their AI models, Reuters reported on 13 October 2026, citing two people familiar with the discussions. The companies were concerned that findings could reveal trade secrets, the people said [1].
What the companies say
Anthropic told Reuters that seeking significant control over embedded evaluators’ publications “did not accurately characterize all of its discussions and work with independent evaluators.” It said that, with evaluators it had worked with before, it had made clear the reviewers “control their methods, evaluation criteria, conclusions and what to publish.” For new evaluator firms, it said it reserved “the right to comment on and approve reports” while they learn how its models work [1].
OpenAI told Reuters it “was not seeking approval rights over the full text of reports produced by evaluators.” It said its standard practice is to redact confidential information, and that evaluators can disclose whether major redactions were made [1].
(Reuters reported both responses in its own words rather than as direct quotes; the wording above is Reuters’.)
What is being negotiated
Outside evaluators test AI models for risks. In “embedded evaluation”, their staff work inside the AI company. Reuters’ sources say the companies worry these staff could take secrets to rivals. Ideas under discussion include having company employees watch embedded teams, and stopping evaluators at one lab from working at another at the same time [1].
Last month Anthropic CEO Dario Amodei said outside evaluators should be able to publish key findings “without editorial control.” OpenAI said days later that assessors should have “editorial independence” [1].
According to the two people, as recently as 5 October both companies were seeking strict editorial control over what evaluators could publish. By 8 October, talks between Anthropic and some evaluators appeared to have moved on, one person said [1].
Reuters also says it saw a contract whose language can bar an evaluation firm from publishing confidential information without the company’s consent [1].
Other voices
“Evaluators are only useful if people can trust what they report, so independence and transparency are critical,” said Marc Warner, CEO of Faculty, which Anthropic named as its first embedded evaluator [1].
“Evaluators have this very privileged access, but whether explicitly or implicitly, they know that it comes at a cost,” said Rumman Chowdhury, CEO of Humane Intelligence and former US Science Envoy for AI [1].
More than 200 researchers, including Geoffrey Hinton, signed a public letter last month calling for independent control of findings, broad access to AI labs and protection from retaliation, Reuters notes [1].
What this means
- Independent testing is now expected, but its rules aren’t settled. The White House accord signed last month lists independent evaluation but does not say how it should work, Reuters reports.
- The fight is over the line between secrets and safety. What a lab treats as intellectual property can overlap with what is needed to judge a model’s risk.
- This account is disputed. It rests on unnamed sources, and both companies say it does not fully reflect their position.
Sources
- Reuters (Aditya Soni), “Exclusive-Behind AI industry’s call for independent reviews, a fight over what gets made public”, 13 October 2026, via 93.3 The Drive. https://www.933thedrive.com/2026/10/13/exclusive-behind-ai-industrys-call-for-independent-reviews-a-fight-over-what-gets-made-public/
Related stories
- Hundreds of Scientists, Including Geoffrey Hinton and Yoshua Bengio, Call for an International Body on AI Containment and Incident Reporting
- White House Super Intelligence Force Says AI Companies Must Report Security Incidents and Remedy Harm; Axios Says Enforcement Is Not Spelled Out
- Anthropic Says Claude Acted on Real Websites During Tests, and Has Cut Live Internet From Its Internal Evaluations
