Anthropic said Thursday it is working with the nonprofits METR and SecureBio and holding talks with other organizations, including Hugging Face, as part of a pledge to embed independent evaluators inside the company.

The move follows CEO Dario Amodei's proposal last month to slow the pace of model development. Amodei said evaluators would help "verify" Anthropic's adherence to safety practices, and the company picked Accenture as its first partner days later.

Under a three-tier system, first-tier evaluators can run narrowly scoped risk investigations, second-tier evaluators can publish reports without Anthropic's approval, and third-tier evaluators get employee-like access. METR — Model Evaluation and Threat Research — is third-tier. Six of its employees have been onboarded with access to Anthropic hardware, office space and tools roughly on par with its own risk teams. They will analyze risk levels, incidents and practices, and publish findings. Anthropic is not funding METR's work.

SecureBio is first-tier, so its access is more limited. It is reviewing the chemical and biological weapons section of the risk report Anthropic released in August, and could take on projects with broader access and publication rights later.

It is unclear what level Hugging Face might receive. Anthropic said it is "working toward an ambitious partnership" with Accenture but did not say how that company fits the tiers. Last month Anthropic said it would fund Accenture's work directly "given the importance and urgency of this work," drawing questions on social media about independence.

Anthropic noted there are no standards yet for what embedded evaluators should access or how they should report findings, and no settled funding system. It has said long-term funding should come from pooled or government sources, as it called for in its June Advanced AI Framework.

Source: CNBC