Hugging Face Applies to Join Anthropic's Embedded Evaluator Program

Hugging Face CEO Clement Delangue announced the establishment of the Open Alignment Initiative, led by co-founder Thomas Wolf, and applied to join the "Embedded Evaluator" program proposed by Anthropic. If approved, Hugging Face evaluators will gain employee-level access, allowing them to be on-site long-term to directly inspect model training and safety measures. Delangue stated that the AI alignment problem cannot be solved by a few leading laboratories alone and requires more external researchers to participate. Anthropic has committed to granting third parties relevant permissions, including workspace access, entry permissions, company computers, and tools to access internal risk teams, and allows evaluators to independently publish their conclusions.