Anthropic just handed six outside researchers the kind of internal access that most AI companies won’t give their own communications teams. That’s not a small thing. As reported by TechCrunch, Anthropic is giving employee-level access to a team from METR, a nonprofit that evaluates frontier AI models, to independently assess its models, safety practices, and risk management.
The timing is deliberate. Since late August, reports of AI agents acting outside their instructions have been surfacing every week or two. Each incident adds pressure on AI labs to show that someone other than themselves is watching. Anthropic’s response is to bring external evaluators inside the building, literally. The METR team gets company-issued laptops, office space, and access to internal tools roughly equivalent to what Anthropic’s own risk teams use. Anthropic employees are told to treat them like colleagues working on alignment risk.
The analogy Anthropic is reaching for is banking regulation, specifically the model where government examiners sit inside the largest banks full-time. It’s a reasonable comparison, and it raises the bar for what “independent review” should mean in this industry. Vague third-party audits with limited data access have been the norm. This is structurally different.
What makes this arrangement credible is the publication rights. METR can publish its own findings on Anthropic’s risks, incidents, and practices without Anthropic’s approval. Anthropic can request redactions in narrow, predefined categories such as customer data, attorney-client privileged material, or model architecture details. But METR can publicly note that redactions were made and whether it considers them justified. That’s a meaningful constraint on Anthropic’s ability to manage its own narrative.
This sits at the top of a three-tier evaluator system Anthropic is building. Lower-tier evaluators, like biosecurity nonprofit SecureBio, run narrower reviews and need Anthropic’s sign-off before publishing. The middle tier gets more information and can publish independently. METR is currently the only organization at the top level.
The move follows a September essay by CEO Dario Amodei, where he proposed embedding outside evaluators in AI labs as an early step toward managing the technology’s development. OpenAI quickly made a similar pledge, and both companies, along with Google, Meta, Nvidia, and xAI, have since signed a White House accord committing to independent review. So Anthropic is not alone here, but it is currently the most specific about what that commitment actually looks like in practice.
For developers and founders building on top of these models, that specificity matters. Independent audits with real teeth are what separate credible safety claims from marketing copy.
