AI Labs Push for Tight Editorial Control Over Independent Evaluations

Anthropic and OpenAI are negotiating new contracts that grant them the authority to approve or edit reports generated by independent AI safety evaluators. The companies also plan to restrict these evaluators from working with competing labs to protect trade secrets.

The move stems from a conflict between the necessity for transparent, independent safety assessments of powerful AI models and the labs' requirement to safeguard proprietary technology and intellectual property.

While both companies publicly advocate for editorial independence, their internal contract negotiations reveal a shift toward implementing restrictive oversight and monitoring of external assessment teams.