Over 100 AI Experts Demand True Independence for Safety Evaluators as Companies Embed Auditors

Leading AI researchers have challenged frontier AI companies to guarantee genuine independence for the safety evaluators they are embedding inside their labs, warning that current proposals lack sufficient structural protections to serve as credible oversight. More than 100 scientists and evaluators, including Geoffrey Hinton and Stuart Russell, published an open letter on September 18, 2026, laying out five non-negotiable conditions for meaningful third-party assessment: structural independence from the companies being evaluated, diverse viewpoints among evaluators, transparent publication of findings, full legal protections against retaliation, and unrestricted access to systems and internal staff.