AI Evaluator Forum urges independent oversight for AI safety evaluations

METR: METR is a nonprofit organization specializing in AI evaluation and safety testing. Members of METR signed the public letter emphasizing the need for scientific objectivity and robust access for evaluators of frontier models. The group focuses on credible, independent assessments to inform risk management. Elon Musk: Elon Musk leads SpaceX and has been active in AI discussions as a founder of xAI. He publicly backed the idea of granting deeper access to independent evaluators for frontier models. His support contributes to growing industry momentum around third-party oversight initiatives. Sam Altman: Sam Altman is the CEO of OpenAI, a major developer of frontier AI models. He has publicly supported the proposal for providing third-party evaluators with significantly expanded access to models and processes. His endorsement aligns with similar statements from other tech leaders on enhancing evaluation credibility. David Sacks: David Sacks served as the former AI czar under President Donald Trump. He has joined the President in adamantly opposing government regulation of AI, favoring industry-led approaches instead. His position influences the broader policy environment around third-party evaluations and model safety. Vinh Nguyen: Vinh Nguyen is a senior fellow for AI at the Council on Foreign Relations and former chief AI officer of the National Security Agency. He signed the letter and stressed the importance of independent evaluators to uncover information that could prevent security failures and economic risks from powerful AI labs. Nguyen has advocated for oversight mechanisms that do not rely solely on company self-reporting. Conrad Stosz: Conrad Stosz serves as chair of the AI Evaluator Forum consortium. He coordinated the public letter on behalf of over 100 signatories and has emphasized the need for shared principles and meaningful independent oversight in AI safety. Stosz has discussed how proposals for enhanced evaluator access could improve risk assessment for unreleased systems. Dario Amodei: Dario Amodei is the CEO of Anthropic, a leading AI company developing frontier models. Over the weekend, he proposed granting certain third-party evaluators "employee-like access" to inspect and audit advanced systems and development processes. This suggestion has sparked broader discussion about embedding evaluators with deeper access rights. Donald Trump: Donald Trump is the President of the United States. He and his former AI czar have strongly opposed government efforts to regulate AI development amid ongoing debates about safety and oversight. His administration's stance contrasts with calls from some industry leaders for formal regulatory measures. Satya Nadella: Satya Nadella is the CEO of Microsoft, which invests heavily in AI infrastructure and partnerships. He has expressed public support for proposals to embed evaluators with greater access to advanced AI systems. His stance reflects broader corporate interest in credible safety practices. Geoffrey Hinton: Geoffrey Hinton is a prominent AI researcher and luminary known for his foundational work in neural networks and machine learning. He signed the public letter urging frontier AI companies to support independent third-party evaluations with strong safeguards for objectivity and access. His involvement highlights concerns among leading experts about managing risks from advanced models. AI Evaluator Forum: The AI Evaluator Forum is a consortium of AI safety experts and evaluators focused on promoting independent oversight of frontier AI development. It organized the public letter calling for standardized conditions that enable credible third-party evaluations. The group seeks to establish basic principles for transparency, independence, and protections to help manage AI risks effectively. Stanford University: Stanford University is a major center for AI research and policy studies. Faculty and researchers from the university joined as signatories to the letter calling for transparency, independence, and protections in third-party AI evaluations. Their participation highlights the role of academic institutions in shaping oversight practices. Johns Hopkins University: Johns Hopkins University is a leading research institution with experts participating in AI safety evaluations. Members of the university signed the public letter advocating for robust conditions for independent embedded evaluators. Their involvement underscores academic contributions to standards for AI risk assessment. Policy Stance: President Donald Trump and his former AI czar have opposed government regulation of AI development, favoring approaches that avoid formal mandates on model oversight. Industry Proposal: Anthropic's CEO proposed granting select third-party evaluators access comparable to privileged employees, including internal systems and candid staff communications. Oversight Standards: A coalition of evaluators is pushing for standardized conditions including transparency, independence, and protections against retaliation to make embedded evaluations credible.