---
title: "Anthropic Lets Outside Auditors Work From Inside the Company"
url: https://media.yoora.tech/en/articles/anthropic-metr-employee-level-access
lang: en
published: 2026-10-15T20:22:43+00:00
updated: 2026-10-15T20:22:43+00:00
section: "News"
description: "Six researchers from the nonprofit METR get employee-level access and can publish their findings without Anthropic's sign-off."
---

# Anthropic Lets Outside Auditors Work From Inside the Company

Anthropic says six researchers from METR, a nonprofit that tests frontier AI models, now have access on par with employees. They get company laptops, office space and internal tools close to what Anthropic's own risk teams use, and staff are encouraged to share with them as freely as with a colleague working on alignment.

## The terms that matter

METR will publish its own findings on the company's risks, incidents and practices, and Anthropic can no longer veto publication. There is one narrow exception: the company may demand redactions in a few predefined categories, namely customer data, attorney-client material and details of model architecture. METR is free to say publicly that it thinks a redaction was not justified.

## A three-tier setup

Anthropic is building three levels of evaluators. Narrow reviewers such as the biosecurity nonprofit SecureBio publish only with approval. A middle tier sees more and publishes without it. The top tier, with employee-level access, currently has one member: METR.

Anthropic compares the arrangement to banking, where regulators keep examiners inside the biggest banks. It also fulfills a promise from Dario Amodei's September essay about embedding outside evaluators in AI labs. OpenAI pledged to do the same, and six major companies have signed a White House accord on independent review.