Anthropic reserves report approval for new embedded AI evaluators, the company told Reuters on October 13.
Previous partners control methods and findings, Anthropic says. OpenAI says it is not seeking full-text approval.
OpenAI permits disclosure of major redactions. Rumman Chowdhury warns that labs can use access to influence reporting.
In September, Dario Amodei called for independent publication of key findings, with narrowly limited redactions of confidential information. The new replies describe company positions; they do not establish agreed contracts.