Anthropic Grants METR Employee-Level Access for Independent AI Safety Review

Anthropic Opens Doors to Independent AI Safety Oversight

Anthropic announced Thursday it has granted employee-level access to six researchers from METR, a nonprofit that evaluates frontier AI models, marking an unprecedented move toward independent AI safety review1

1

2

2

. The METR team now operates with Anthropic-issued company laptops, office space, and access to internal tools roughly equivalent to what the company's own risk assessment teams use. This arrangement allows METR to independently assess Anthropic's models, safety practices, and AI risk management without the company having veto power over publications.

Why This Matters for AI Governance

The timing of this announcement is significant. Incidents of AI agents acting outside their instructions have been disclosed every week or two since late August, sharpening calls for independent checks on AI companies1

1

. Anthropic stated it wants to be publicly accountable for its decisions and aims to "close the gap between what evaluators understand about a frontier AI company's systems and what the company itself understands." The company compared the arrangement to banking regulations, where resident examiners are placed inside the largest banks to ensure independent oversight1

1

.

Publication Rights and Redaction Boundaries

METR will publish its own assessments of Anthropic's risks, incidents, and practices, with Anthropic having given up the right to block them. The company can only require redactions for information falling into narrow, pre-defined categories such as customer data, attorney-client privileged information, or model architecture details1

1

2

2

. Critically, METR retains the right to publicly characterize any redactions and state whether it believes they were justified, as long as it does so without revealing the redacted information1

1

.

Three-Tier Evaluator System Takes Shape

Anthropic is building a three-tier system for external evaluation of AI models. METR operates at the third tier with full employee-level access. The first tier includes evaluators like biosecurity nonprofit SecureBio, which runs narrow reviews and requires Anthropic's approval to publish. SecureBio is currently reviewing the chemical and biological weapons chapter of Anthropic's August Risk Report3

3

. The middle tier receives more information and can publish without approval, while the top tier adds employee-like access1

1

. Anthropic is also working with Accenture's AI business Faculty as evaluators, with partnerships covering AI model evaluation, adversarial testing, and assessments of whether models behave as intended3

3

.

Industry-Wide Implications and White House Accord

Thursday's announcement fulfills a promise Anthropic CEO Dario Amodei made in a September essay, where he proposed embedding outside evaluators inside AI labs as a first step toward slowing the technology's development1

1

. OpenAI quickly pledged the same approach, and both companies, along with Google, Meta, Nvidia, and xAI, have since signed a White House Accord committing to independent review1

1

. The outcome of these arrangements will help determine how the world scrutinizes a technology that some of its creators warn may be advancing beyond humanity's ability to manage3

3

.

Balancing Trade Secrets and Public Accountability

Reuters exclusively reported earlier this week that Anthropic and OpenAI had sought significant control over what outside reviewers could publish, concerned that findings could leak trade secrets3

3

. A person familiar with the matter told Reuters that talks with Anthropic had evolved recently, with the company becoming more open in negotiations. A key sticking point between evaluators and AI labs is that information companies might consider crucial intellectual property can also be relevant to assessing a model's risks, leading to disputes and redactions3

3

. Experts have also raised concerns about the independence of evaluators, as Anthropic and OpenAI choose who gets to carry out the reviews3

3

.

What to Watch For

Watch for METR's first published assessments under this new arrangement, which will test the practical boundaries of the redaction policy and set precedents for independent AI safety review. The expansion of SecureBio's access and publication rights after completing its current review will signal whether Anthropic's tiered system can scale effectively. Also monitor whether other major AI labs follow through on their White House Accord commitments with similar employee-level access arrangements, and whether the independence concerns raised by experts materialize as evaluators begin publishing findings.