Oct 15, 2026AnnouncementsAn update on independent evaluators

Anthropic Embeds Independent Auditors with 'Employee-Like' Access

Anthropic announced it is granting deep, internal access to independent evaluators, starting with the nonprofit METR, to assess the risks of its frontier AI models. The move represents a significant step toward external oversight in an industry often criticized for its lack of transparency. By providing a third party with access comparable to its own risk teams, Anthropic is attempting to establish a new standard for accountability that goes beyond corporate pledges and internal safety reports.

A Tiered Framework for Verification

The company has established a three-tiered system for evaluator access. METR is the first organization to operate at Tier 3, the highest level, which provides extensive access and publication rights. This partnership is designed to close the gap between what an AI company knows about its own systems and what an external auditor can verify. Anthropic is also working with SecureBio and Accenture in earlier-stage partnerships.

Key details of the Tier 3 METR engagement include:

• Six METR employees have been onboarded with Anthropic hardware, office space, and access to internal tools.
• METR will publish its own risk assessments and analysis of Anthropic's safety practices.
• Anthropic cannot veto publication but can request redactions for narrow categories of sensitive information, such as customer data or model architecture.
• METR retains the right to publicly state whether it believes any redactions were justified.

Impact on Industry Oversight

This model for embedded evaluation directly addresses the challenge of verifying AI safety claims without forcing companies to publicly disclose proprietary or dangerous information. By creating a trusted intermediary framework, similar to how resident examiners operate within major banks, Anthropic is setting a practical precedent for the AI industry. The initiative puts pressure on other frontier AI labs to move from abstract commitments, like the White House Accord on Super Intelligence, to implementing concrete, verifiable oversight mechanisms.

By operationalizing a tiered audit system with METR at its core, Anthropic is not just fulfilling a pledge but actively building a regulatory-style oversight model that challenges the industry's default posture of opacity, effectively forcing competitors to demonstrate, rather than just declare, their commitment to safety.