Anthropic Gives METR Employee-Like Access Under a New Three-Tier Independent Evaluator Framework
TL;DR: Anthropic announced on October 15, 2026 that it has begun giving a team of six METR employees employee-like access to its internal systems so the nonprofit can independently assess the risks of its frontier models and publish the results. The arrangement is the first concrete instance of Anthropic's new three-tier framework for third-party evaluators, and Anthropic says it is embedding evaluators voluntarily and believes every frontier AI company should do the same.
What Anthropic announced
Anthropic has started giving employee-like access to a team from METR, a nonprofit that evaluates frontier AI models. METR will publish the results of its risk assessments.
The announcement follows two prior commitments. Anthropic says that "last month" it committed to stronger independent, third-party assessment of its models and safety practices, and that it also signed the White House Accord on Super Intelligence, in which companies pledged to partner with independent external auditors or evaluators.
Anthropic states that it has worked with METR over the past year on several reviews of its Risk Reports. Beyond METR, Anthropic says it is working with SecureBio and Accenture as evaluators, and is in early discussions with several other organizations, including Hugging Face.
Why Anthropic says independent evaluators are necessary
Anthropic's stated reason for third-party evaluation is that it cannot publicly share all the information relevant to its safety work. Some of that information could be useful to malicious actors, and some is commercially sensitive, such as compute or training information.
Instead of publishing that information, Anthropic says it is giving third-party evaluators deep access and having them share evidence and reasoned conclusions about the risks involved in building its models. The stated goal is to close the gap between what evaluators understand about a frontier AI company's systems and what the company itself understands.
Anthropic lists four requirements an evaluator needs to assess frontier models:
- Expertise in assessing the capabilities of frontier models, including understanding the way evaluations can go wrong, such as alignment faking.
- Independence: evaluators should not hold equity in the company they are evaluating, or be financially or operationally dependent on it.
- Access sufficient to meaningfully evaluate risks. Beyond access to the latest models, evaluators need visibility into how those models were trained, into how safeguards function, and more.
- Freedom to publish observations and assessments of a model or a process.
The bank supervision analogy
Anthropic compares its approach to bank supervision, while noting the analogy is not exact. Banks know more about their own risk positions than outsiders do, so regulators assign resident examiners to work inside the largest banks. If a bank's risk-taking could threaten the public, third parties try to see that in advance.
Anthropic frames embedded evaluation for the AI industry as a step toward robust safety oversight, and says it is moving quickly with multiple evaluators and learning actively about what kinds of partnership work.
What "employee-like access" means at Anthropic
Anthropic has onboarded six METR employees with company hardware, office space, and access to internal tools and systems that are roughly on par with what Anthropic's own risk assessment teams have access to.
Outside a small number of clearly defined categories — Anthropic names data about customers, attorney-client privileged information, and information about its model architectures — Anthropic employees are encouraged to share information with the embedded METR team just as they would with a fellow employee working on alignment risk.
Anthropic states that its engagement with METR is limited to a small team operating under enforceable agreements. Other METR employees are siloed from that team's work, including researchers who work with Anthropic on other projects. Anthropic describes embedded evaluation as a way for a few qualified people working under Anthropic's security standards to get a close view of how the company operates, and explicitly not a way for other organizations to obtain Anthropic's information or that of its customers and partners.
Publication rights and redaction
METR will publish its analysis of Anthropic's risk levels, incidents, or practices. Anthropic cannot veto publication, but it can require METR to redact information if it falls into narrow, predefined categories such as customer data, attorney-client privileged information, or model architecture information.
METR is free to characterize Anthropic's redactions publicly and to say whether it believes the redactions were justified, as long as it does so without revealing the redacted information.
The three tiers of evaluator access
Anthropic says it is developing three tiers of access for evaluators, and that over time it hopes to have several evaluators at the third tier, with each team publishing its own analysis of Anthropic's risk posture and key decisions.
| Tier | Access | Publication rights |
|---|---|---|
| First tier | Anthropic decides what access to provide | Evaluators can publish findings with Anthropic's approval |
| Second tier | Access to more information | Evaluators can publish without Anthropic's approval; Anthropic can still require removal of specific categories of information, and evaluators can publish their views on whether those redactions were justified |
| Third tier | Employee-like access | Publication rights similar to the second tier |
The arrangement with METR is the third tier.
Other evaluator partnerships
Because Anthropic's other partnerships are still in early stages, those evaluators receive more limited information and have more limited publication rights. Anthropic says the goal of these early engagements is to make sure partners protect confidential information and have the requisite resources and expertise to evaluate its models, and that it hopes to fully embed more independent evaluators as partnerships progress and trust is built.
SecureBio is currently in the first tier. It is in the process of reviewing the chemical and biological weapons chapter of Anthropic's August Risk Report, after completing and publishing a pilot review of a previous risk report. Anthropic says it will consider moving to projects with increased access and publication rights after this review is complete.
Accenture and Faculty are working with Anthropic toward what Anthropic describes as an ambitious partnership that includes evaluating and red-teaming models, conducting alignment assessments, testing model safeguards, and building capabilities across the organizations.
Hugging Face is named among the organizations with which Anthropic says it is in early discussions.
What this means for frontier AI oversight
Anthropic's announcement establishes a concrete, tiered mechanism for outside organizations to inspect a frontier lab's safety practices without requiring the lab to publish sensitive details. The key structural features are that publication rights are guaranteed rather than discretionary at the higher tiers, that redaction is limited to predefined categories, and that evaluators may publicly dispute whether redactions were justified.
Anthropic states that it is embedding evaluators voluntarily, and that it believes every frontier AI company should do the same. The company says it is committed to sharing updates as it goes.