Anthropic and OpenAI embed safety evaluators, but will they be independent
anthropic and openai plan to embed third-party evaluators to report safety incidents and assess model alignment, but details on independence and access remain unclear. this affects the ai industry's safety standards.
anthropic ceo dario amodei proposed embedding third-party evaluators inside frontier ai companies, giving them access to report safety incidents and assess model alignment. openai ceo sam altman also committed to the practice. evaluators welcome the proposal but stress that details, potentially backed by legislation, are needed to ensure independence. this development affects the ai industry's safety standards and transparency.