AI Evaluator Forum Open Letter Sets Minimum Conditions for Third-Party Risk Assessments

The AI Evaluator Forum published an open letter on September 18, 2026, setting minimum conditions for third-party risk assessments. The letter states that AI labs such as Anthropic and OpenAI must meet specific requirements to effectively cooperate with external evaluation organizations.

The letter was co-signed by dozens of AI figures, including computer science pioneers Geoffrey Hinton and Stuart Russell.

According to the letter, on-site third-party evaluators must maintain scientific objectivity, transparency, and independence, and be protected from interference by the companies being evaluated. Evaluators should be allowed to work openly, communicate directly with company boards and other oversight bodies without filtering or restriction, and publicly release their findings.

The letter also calls for evaluators to be protected from retaliation for their work, and to receive the same access to systems, data, tools, and physical premises as internal evaluators. In addition, each AI lab should bring in evaluators with diverse perspectives and backgrounds to cover risks such as bioscience risks and risks of AI losing control.

Earlier, Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman both pledged to give third-party evaluators access to internal resources and information at the same level as employees. The open letter responds to those commitments. Earlier this week, Microsoft CEO Satya Nadella also said he welcomed the idea of having third-party evaluators embedded in AI companies.