Anthropic Partners with METR and SecureBio for Independent AI Risk Evaluation
October 22, 2026 · 3 min read
Anthropic has established a partnership with the nonprofit organizations METR and SecureBio to evaluate the company's artificial intelligence risks and security practices. This collaboration was announced on Thursday, October 15, 2026, as part of a broader initiative to integrate independent evaluators into the development process. The move follows a proposal by Anthropic's chief executive officer, Dario Amodei, published the previous month, which advocated for slowing the pace of model development to address safety concerns.
The partnership structure relies on a three-tier system of evaluators, each with distinct levels of access to Anthropic's internal systems. First-level evaluators conduct risk investigations with a restricted scope. Second-level evaluators are permitted to publish reports without requiring prior approval from Anthropic. Third-level evaluators possess access rights equivalent to those of internal employees. METR is classified under the third level, with six of its professionals integrated into Anthropic's teams. These individuals have access to equipment, offices, and tools comparable to internal risk evaluation staff. They will analyze risk levels, incidents, and company practices, publishing their conclusions without financial support from Anthropic for their work.
SecureBio is currently reviewing the section of Anthropic's August 2026 risk report that addresses chemical and biological weapons. The organization may expand its involvement in projects with greater access and publication rights in the future. Anthropic is also in negotiations with other organizations, including Hugging Face, to expand the program of embedded independent evaluators. The level of access to be granted to Hugging Face has not yet been defined. Hugging Face, known for its open-source development platform, was acquired by Nvidia for approximately 13 billion US dollars in the month prior to the announcement.
The role of Accenture remains unresolved. Anthropic stated it is working toward an ambitious partnership with the consultancy but has not disclosed which level of the system it will occupy. Previously, the company indicated it would directly fund Accenture's work, a decision that generated criticism on social media regarding the consultancy's independence. Anthropic acknowledged the lack of established standards for funding independent evaluations. The company stated that long-term funding should come from collective or governmental sources, as advocated in its Advanced AI Framework in June. Since such sources do not currently exist, Anthropic plans to work with different evaluators under varying funding arrangements.
Independent evaluators like METR previously occupied a discreet space in the AI industry. Their relevance has grown amid increasing concerns about the ability of AI companies to protect their models. In late September 2026, the President of the United States, Donald Trump, encouraged industry companies to associate with an independent external auditor or evaluator in a voluntary agreement. Signatories of this agreement include executives from Google, Meta, OpenAI, SpaceX, and Nvidia. Anthropic stated that it is incorporating evaluators voluntarily and believes all frontier AI companies should do the same.