Artificial Intelligence
Anthropic gives METR employee-like access for model checks
Anthropic said it had started giving METR employee-like access so the nonprofit could assess its models and safety practices, while it also described other evaluator partnerships and a three-tier access system.
- Anthropic said it had started giving a METR team employee-like access.
- METR was described as a nonprofit that evaluates frontier AI models and will publish its risk assessments.
- Anthropic said it had onboarded six METR employees with company hardware, office space and access to internal tools and systems.
- The company said SecureBio was in the first tier of its evaluator programme, while it was working with Accenture and Faculty on a broader partnership.
- Anthropic said it was in early discussions with other organisations, including Hugging Face.
Anthropic said it had begun giving a team from METR employee-like access as part of its approach to independent review. It said the move followed a commitment to stronger third-party assessment of its models and safety practices and its signing of the White House Accord on Super Intelligence.
The company said METR would publish its risk assessments and that the arrangement gave the nonprofit access to hardware, office space and internal tools. Anthropic said it would not be able to veto publication, though it could require redaction of narrow categories such as customer data, attorney-client privileged material and model architecture information.
Anthropic said it had worked with METR over the past year on reviews of its Risk Reports. It also said evaluators needed expertise, independence, enough access to judge risks and freedom to publish observations.
The company said it was developing three tiers of access for evaluators, with SecureBio in the first tier and Accenture and Faculty involved in a wider partnership. It said it was also in early discussions with several other organisations, including Hugging Face.
Named in this story
Companies
- Anthropic
- announced the evaluator access arrangements and its tiered oversight model
- Accenture
- was said to be working with Anthropic on evaluator and red-teaming partnerships
- Faculty
- was named by Anthropic as part of the partnership work toward evaluation and red-teaming
- Hugging Face
- was said to be in early discussions with Anthropic as an evaluator
Organisations
- METR
- received employee-like access and will publish its assessments
- SecureBio
- was said to be working with Anthropic as a first-tier evaluator
Governments and agencies
- White House Accord on Super Intelligence
- the accord Anthropic said it had signed with other companies
How the source tells it
The piece was a corporate announcement written in a measured, self-justifying register, with some reassurance built through repeated claims about independence, oversight and safeguards.
- reassurance it repeatedly framed the arrangement as strengthening independent oversight and limiting access to sensitive information
- self-promotion it presented its partnership model as a template other frontier AI companies should follow