On October 15, Anthropic announced that it has begun giving "employee-like" access to six researchers from METR, a nonprofit that evaluates frontier AI models. The goal is for outside evaluators to examine the company's risks, incidents and practices from the inside, and to be able to publish their conclusions. The company says it is doing this voluntarily and that every frontier AI company should do the same.

What Anthropic announced

According to the official post, the METR researchers receive hardware, office space, and internal tools and systems, broadly comparable to those of Anthropic's own risk assessment teams. The company encourages sharing information with them, except for customer data, attorney-client privileged material and information about model architecture. Only a small, contractually bound team has this access; the rest of METR is kept separate from the work.

On publication, Anthropic says METR will release its analysis of the company's risk levels, incidents or practices. Anthropic cannot veto publication, but it can require the removal of predefined categories of information, and the company decides what falls into those categories. METR can state what level of access and information it received.

The three evaluator tiers

  • Tier 1: narrowly scoped risk investigations. Anthropic sets the access, and findings are published only with its approval.
  • Tier 2: evaluators get more information and can publish without approval. Anthropic can still require certain categories to be removed, but evaluators can say whether they think the redactions were justified.
  • Tier 3: this is the METR arrangement, with employee-like access and publication rights similar to tier 2.

The company names other partners: SecureBio is reviewing the chemical and biological weapons chapter of the August 2026 Risk Report, and Accenture and Faculty are working toward a broader partnership that includes model evaluation and red-teaming. There are early conversations with other organizations, including Hugging Face. Once SecureBio finishes, Anthropic says it will consider projects with more access and hopes to have several tier 3 evaluators, each publishing its own analysis.

Criticism and what is still unknown

DNYUZ recalls that the commitment was announced by Dario Amodei in a September essay, and that the names of the six researchers have not been disclosed. METR did not respond to the outlet's request for comment. Officechai reports that METR's independence was questioned by an X user who pointed to financial links between the organization and Anthropic. That point is a third-party allegation, and Portal GTA 6 has not independently verified it.

DNYUZ also says Sam Altman stated that OpenAI would do something similar, but that this has not happened yet. Anthropic says there are no common standards for what embedded evaluators can see or how they should report. In short, it is a concrete step toward transparency, but proof that it works will depend on what METR actually publishes.