Anthropic and OpenAI sought control over what outside safety evaluators may publish, Reuters reports, weeks after promising them independence

Reuters reported on 13 October, in an exclusive by Aditya Soni, that Anthropic and OpenAI have sought significant control over what outside evaluators can publish about their models, citing trade secrets. Two people familiar with the talks told Reuters that both labs were still seeking strict editorial control as of 5 October. One of them said Anthropic's position had shifted in talks with some evaluators by 8 October.

The report sets that against the labs' own words. “Anthropic CEO Dario Amodei said last month 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.”” Independent evaluation is one of four safety measures in the White House Accord on Super Intelligence, signed in late September, but Reuters says the accord does not set out how such reviews should work.

Reuters says standard language in some evaluator contracts can bar the firms from publishing confidential information without the companies' consent. The labs worry that evaluators embedded inside them could carry secrets to rivals; options discussed include lab staff observing embedded teams as they work, and keeping an evaluator embedded at one lab from working at another at the same time.

Both labs pushed back. Anthropic told Reuters it reserves the right to comment on and approve reports only for new evaluator firms. OpenAI said it does not seek approval of full reports, redacts confidential information and lets evaluators disclose major redactions. No contract was visible from here.

• Reported Two people familiar told Reuters both labs were seeking strict editorial control over evaluators' publications as of 5 October. Reuters, via Lufkin Daily News
• Reported Reuters says language in some evaluator contracts can bar publishing confidential information without the companies' consent. Reuters, via Lufkin Daily News
• Claimed Anthropic says it reserves comment and approval rights only for new evaluator firms; OpenAI says it does not seek approval of full reports. Reuters, via Lufkin Daily News