Experts call for independent AI safety evaluators for Anthropic and OpenAI

More than 100 AI experts and evaluators have signed a public letter urging frontier AI companies, including Anthropic and OpenAI, to give thirdparty testers the access and protections needed to assess model safety. The letter, published Friday and shared exclusively with CNBC, says independent evaluators should have scientific objectivity, transparency, and freedom from interference when reviewing AI systems, training practices, deployment decisions, and major incidents. The group argues that as advanced AI models become more powerful, companies should not be the only source of information about safety and security risks. Signatories include Geoffrey Hinton and researchers from institutions such as Stanford, Johns Hopkins, and the nonprofit METR. The letter also calls for clearer standards on evaluator independence, access to internal systems, transparency about findings, and protections against retaliation from the companies being assessed.