Independent safety evaluators urge Anthropic and OpenAI for greater access

Over 100 AI experts and evaluators are urging Anthropic, OpenAI, and other foundation model labs to grant independent safety testers greater access, transparency, and protections, warning current conditions are insufficient for credible oversight. The coalition, organized by the AI Evaluator Forum, published a public letter Friday demanding third-party evaluators receive "employee-like access" to systems and data, be shielded from retaliation, and maintain full editorial control. Signatories include Geoffrey Hinton and researchers from Stanford and Johns Hopkins. The call follows Anthropic CEO Dario Amodei's weekend proposal for embedded evaluators, which OpenAI's Sam Altman and others have publicly supported. The group says independent oversight is critical as frontier models pose potential risks to cybersecurity and national infrastructure.