Anthropic taps nonprofits METR and SecureBio to help assess risks and safety practices
The company is also engaged in early-stage discussions with other organizations, including Hugging Face.
Key points
- Anthropic announced partnerships with nonprofits METR and SecureBio and talks with Hugging Face to embed independent evaluators.
- METR is a third-tier evaluator with employee-like access; SecureBio is first-tier with limited access.
- Anthropic said it is not funding METR's work and is working toward a partnership with Accenture.
Anthropic announced on Thursday that it is working with the nonprofits METR and SecureBio, and holding discussions with other organizations including Hugging Face, as part of its commitment to embed independent evaluators within the company.
The AI lab had pledged to work closely with third-party evaluators after CEO Dario Amodei published a proposal last month to slow the pace of model development. Amodei said the evaluators would help "verify" Anthropic's adherence to safety practices, and the company selected Accenture as its first partner days later.
Tiered access system
Anthropic is developing three tiers for evaluators with different levels of access. First-tier evaluators can conduct narrowly scoped risk investigations, second-tier evaluators can publish reports without Anthropic's approval, and third-tier evaluators have employee-like access, according to a release.
METR, which stands for Model Evaluation and Threat Research, is considered a third-tier evaluator. Anthropic said it has onboarded six METR employees who have access to Anthropic hardware, office spaces and tools that are "roughly on par" with its own risk assessment teams. They will analyze Anthropic's risk levels, incidents and practices, and publish their findings. Anthropic said it is not funding METR's work.
We're embedding evaluators voluntarily, and we believe every frontier AI company should do the same.
— Anthropic
SecureBio and other partners
SecureBio is a first-tier evaluator, so its access is more limited. It is reviewing the chemical and biological weapons section of the risk report Anthropic released in August. Anthropic said SecureBio could participate in projects with broader access and publication rights in the future.
It is not clear what level of access Hugging Face could be granted. Anthropic said it is working toward an ambitious partnership with Accenture but did not disclose how that company fits into the tiered system. Anthropic said last month it would fund Accenture's work directly, prompting critics to raise concerns about its independence on social media.
