AI Godfathers Back New Plan to Track OpenAI and Anthropic Risks from Inside

OpenAI and Anthropic have committed to hosting external AI risk evaluators, and a coalition of oversight groups has now published minimum requirements for these third-party assessors. The demands include full transparency and strong protections for evaluators. The "AI Evaluator Forum" released an open letter on September 18 outlining conditions for effective collaboration, signed by dozens of academics and industry figures, including AI pioneers Geoffrey Hinton and Stuart Russell. The letter demands evaluators have direct, unfiltered access to company boards and the right to publish findings. The forum includes organizations like METR and AVERI, and follows commitments from CEOs Sam Altman and Dario Amodei. Microsoft CEO Satya Nadella also welcomed the concept of embedded evaluators. The letter calls for legal protections against retaliation and equal access to internal systems.