METR Gets Employee-Like Access at Anthropic: An Update on Independent Evaluators!

Hello, it's Shii-chan! Today I'm covering an Anthropic News post about independent evaluators.

Anthropic News anthropic.com

What was announced?

In the Anthropic News post An update on independent evaluators (October 15, 2026), Anthropic says it has begun giving employee-like access to a team from METR, a nonprofit that evaluates frontier AI models. The post presents this as a follow-up to its commitment to stronger independent, third-party assessment and to its signing of the White House Accord on Super Intelligence.

Why it matters

Anthropic writes that it wants public accountability but cannot share all safety information, since some could help malicious actors or is commercially sensitive.

It compares the approach to bank regulators placing resident examiners inside large banks, while noting the analogy is imperfect.

What changes

Six METR employees have company hardware, office space, and internal tools roughly on par with Anthropic's own risk assessment teams. METR will publish its analysis of risk levels, incidents, or practices.

Dive Deep

Here are the publication rules, evaluator criteria, and access tiers.

Publication:

• Anthropic cannot veto publication
• It can require redactions in narrow, predefined categories: customer data, attorney-client privileged information, and model architecture information
• METR may publicly say whether it thinks the redactions were justified, without revealing the redacted content

Evaluator criteria:

• Expertise in frontier model capabilities
• Independence (no equity or financial or operational dependence on Anthropic)
• Enough access to assess risks
• Freedom to publish

Three access tiers:

• Tier one: narrow investigations, with Anthropic approving publication
• Tier two: publication without approval, and broader information access
• Tier three: the METR arrangement, with employee-like access and publication rights similar to tier two

Other partners: SecureBio is in the first tier and is reviewing the chemical and biological weapons chapter of Anthropic's August 2026 Risk Report, after publishing a pilot review of an earlier report. Anthropic is also working with Accenture and Faculty on a broader partnership covering red-teaming, alignment assessments, and safeguard testing, and is in early talks with others, including Hugging Face.

Because evaluators may work with other companies, Anthropic says it is careful about inadvertent leaks, especially of architecture information.

Wrap-up

• Six METR employees now have employee-like access
• Anthropic can't veto publication, and redactions are limited to narrow categories
• There are three access tiers, and METR is in tier three
• Work with SecureBio, Accenture, Faculty, and others is underway
• Anthropic says every frontier AI company should do the same

This one is for anyone interested in AI safety and third-party evaluation!