Geoffrey Hinton Co-Signs Open Letter Demanding Independent Third-Party AI Safety Assessments
A coalition of watchdog organizations, co-signed by AI pioneers Geoffrey Hinton and Stuart Russell, published an open letter establishing baseline requirements for third-party AI safety evaluators. The initiative responds to commitments from Anthropic and OpenAI to allow external evaluators, insisting that third-party experts must operate free from corporate intervention. As frontier AI models rapidly advance, this open letter highlights concerns that self-regulation and voluntary safety commitments could become public relations exercises. Establishing binding protections, direct board reporting, and unmonitored access for external auditors is critical to ensuring effective AI governance and transparency. The letter stipulates that embedded evaluators must have equal system and data access as internal teams, full authority to publish findings, and direct communication lines to company boards without corporate filtering. Additionally, it demands safeguards against retaliation from AI labs and calls for diverse evaluation teams spanning biosecurity and loss-of-control risk disciplines.
## BACKGROUND
Frontier AI developers like OpenAI and Anthropic produce increasingly capable models that carry complex risk profiles, including potential misuse for cyberattacks or biological threats. Third-party evaluation organizations conduct red-teaming and safety audits on these models, but their ability to identify flaws depends heavily on receiving unfiltered access and remaining independent from commercial incentives.