PolicyTechCrunch

Anthropic and OpenAI want to embed safety evaluators. Will they really be independent?

#ai safety#independent evaluators#anthropic#openai#policy

English

Anthropic and OpenAI have proposed embedding independent safety evaluators within their companies to assess AI models' alignment and safety incidents. While this initiative has been welcomed by third-party evaluators, concerns remain about the true independence of these evaluators and whether AI companies will allow sufficient access to their systems for meaningful assessments.

中文

Anthropic 和 OpenAI 提出了在公司内部嵌入独立安全评估者的建议，以评估 AI 模型的对齐性和安全事件。虽然这一举措得到了第三方评估者的欢迎，但仍然存在关于这些评估者真正独立性以及 AI 公司是否会允许足够的系统访问以进行有意义评估的担忧。