Anthropic has brought in nonprofit evaluators METR and SecureBio, and is talking to others such as Hugging Face, as it expands external oversight of its safety practices.
01Summary
- 01Anthropic said on Thursday it is working with METR and SecureBio as independent evaluators, alongside talks with other organisations including Hugging Face.
- 02METR, or Model Evaluation and Threat Research, is a third-tier evaluator and has six employees embedded at Anthropic with access to hardware, office space and tools.
- 03SecureBio is a first-tier evaluator and is reviewing the chemical and biological weapons section of Anthropic's risk report released in August.
- 04Anthropic said it is developing three tiers of evaluator access, ranging from narrow risk investigations to employee-like access and publication rights.
- 05CEO Dario Amodei proposed slowing model development last month, and Anthropic later selected Accenture as its first partner for evaluator work.
02Why it matters
The move shows how one of the best-known AI firms is trying to formalise outside scrutiny of its systems. It may shape how other frontier AI companies handle safety checks and public trust.
Pulse AIAsk about this story
Answers are generated by AI from the story summary; they are not investment, health or legal advice.
