Over 100 AI Experts Urge Independent Safety Evaluators for Frontier Models
More than 100 artificial intelligence experts and evaluators have signed a public letter warning that they lack the necessary resources and protections to effectively test the safety of rapidly advancing AI models. The group, organized by the AI Evaluator Forum consortium under chair Conrad Stosz, is calling for greater scientific objectivity, transparency, independence, and robust safeguards in third-party evaluations. The initiative follows Anthropic CEO Dario Amodei's proposal offering evaluators more access to inspect cutting-edge foundation models. While some industry leaders like Sam Altman of OpenAI support the idea, logistical challenges remain unresolved regarding evaluator selection and depth of inspection into sensitive technologies. The signatories emphasize that independent oversight is essential for managing AI risks but do not advocate a specific regulatory approach. The letter highlights growing concerns about frontier model safety as companies increasingly call for third-party evaluations to address potential dangers. This effort aims to ensure accountability from major foundation model providers and establish standardized principles for evaluators working outside the primary labs.
Topics: Artificial Intelligence Regulation Safety Testing Third-party Evaluation