A coalition of AI oversight organizations, the AI Evaluation Forum, has issued an open letter outlining mandatory conditions for third-party risk assessments. The document specifies that AI labs, including OpenAI and Anthropic, must provide external evaluators with the same level of access to systems, data, and tools as internal staff to ensure effective collaboration.
The letter, signed by industry figures such as Geoffrey Hinton and Stuart Russell, demands that evaluators maintain scientific objectivity and independence. The forum insists that these experts must be protected from corporate retaliation and given the authority to communicate directly with company boards and oversight bodies without interference from the AI labs themselves.
Furthermore, the proposal calls for transparency, granting evaluators the right to publish their findings publicly. The forum also suggests that labs should recruit experts with diverse professional backgrounds to address specific threats, such as biological risks and potential loss of AI control. Microsoft CEO Satya Nadella has expressed support for the concept of third-party on-site assessments.
These requirements follow recent commitments from OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei, who previously pledged to grant third-party auditors access to internal resources. The open letter serves as a formal response to these promises, aiming to establish a standardized framework for independent oversight within the rapidly evolving artificial intelligence sector.
