Anthropic announced on Oct 15, 2026 that it has begun giving a team from METR — a nonprofit that evaluates frontier AI models — access "roughly on par" with its own risk-assessment staff. The six onboarded evaluators get company hardware, office space and internal tools, with only narrow exclusions such as customer data, attorney-client privileged material and information about model architectures. METR can publish its analysis of Anthropic's risks and incidents, and Anthropic cannot veto publication, though it may require redactions in predefined categories. The post describes a three-tier access model and says Anthropic is also working with SecureBio, which is reviewing the chemical and biological weapons chapter of its August Risk Report, and Accenture, while holding early talks with others including Hugging Face. Anthropic says it is embedding evaluators voluntarily and argues every frontier AI company should do the same.
Our takeAnthropic frames third-party evaluation as the AI industry's version of resident bank examiners: outsiders with insider access. The logic is genuinely sound — the safety claims a lab can't publish are exactly the ones that most need outside verification. The catch is that Anthropic still defines the redaction categories and chooses who gets deep access, so independence leans on the company's own restraint. Whether this becomes industry norm or a well-publicized moat depends on whether rivals follow.