Anthropic gives METR employee-like access and veto-free publication rights

Anthropic’s post says that last month it “committed to stronger independent, third-party assessment” of its models and safety practices, and signed the White House accord in which companies pledged to work with independent auditors or evaluators. METR, which has reviewed Anthropic’s Risk Reports several times, is the first evaluator with “employee-like access”. Six METR staff now work inside Anthropic with its tools and systems.

The key term is publication. METR will publish its analysis of risk levels, incidents or practices; Anthropic cannot veto it and can only require redaction of narrow, predefined categories. METR may then say publicly whether it thinks each redaction was justified. Two days earlier Reuters had reported that labs sought control over what evaluators publish, which makes this term the main point of the deal.

Anthropic set out three tiers of access. SecureBio sits in the lowest one: it is reviewing the chemical and biological weapons chapter of the August Risk Report, and publication needs Anthropic’s approval. Accenture, with Faculty, does evaluation and red-teaming work. Hugging Face is in early talks.

This is the first public case of a frontier lab placing outside evaluators inside the company with publication rights it cannot veto. Amodei’s “pace the frontier” essay proposed it as a first step. It came the same day the FTC sent civil investigative demands to Anthropic and OpenAI over safety incidents. Neither the post nor the coverage we read includes a statement from METR itself.