Artificial Intelligence

Anthropic Grants AI Watchdog METR Employee-Like Access

The artificial intelligence lab has embedded external evaluators from the nonprofit METR inside its offices, granting them hardware, internal tools, and the freedom to publish findings without company veto.

Radwan AbdulhadiEditor and technology writer

Published: October 18, 2026 · Updated: October 18, 2026 · 3 min read

Anthropic Grants AI Watchdog METR Employee-Like Access — Image: Ars Technica

Providing Employee-Level Access to External Auditors

Anthropic has officially granted six researchers from the independent nonprofit [METR](https://metr.org) “employee-like access” to evaluate its artificial intelligence models, internal safety practices, and overall risk management. As detailed in a company [blog post](https://www.anthropic.com/news/an-update-on-independent-evaluators), the embedded team has received company-issued laptops, dedicated office space, and access to internal systems roughly on par with what Anthropic’s own internal risk assessment teams utilize.

The arrangement represents a notable escalation in third-party safety oversight for frontier artificial intelligence developers. According to the company, Anthropic employees have been actively encouraged to share internal information with the visiting evaluators just as they would with a fellow colleague working specifically on alignment risk.

Independent Publication Rights and Redaction Limits

A central component of the new oversight framework is that METR will independently publish its own assessments covering Anthropic’s model risks, operational incidents, and safety practices. Significantly, Anthropic has given up its right to block the publication of these independent findings.

While Anthropic cannot veto publication altogether, the company retains the ability to require the redaction of information falling into narrow, pre-defined categories. These exempted areas include customer data, information protected by attorney-client privilege, and technical details regarding specific model architectures. Furthermore, METR remains free to publicly characterize these redactions and state whether it believes the exclusions were justified, provided it does so without disclosing the underlying protected content.

Anthropic is letting METR researchers look under the hood with employee-level access — Image related to the report from TechCrunch · Source: TechCrunch

Fulfilling Pledges Made by Frontier AI Leadership

The decision to embed external auditors directly follows a proposal set forth by Anthropic CEO Dario Amodei [in a September essay](https://darioamodei.com/post/we-must-pace-the-frontier), where he advocated for placing independent evaluators inside major AI laboratories as a vital step toward better pacing and monitoring the technology's rapid development. The proposal quickly garnered support across the industry, with [OpenAI quickly pledged the same](https://techcrunch.com/2026/09/12/anthropic-ceo-outlines-plan-to-pace-the-frontier/) following the initial announcement.

Anthropic has structured its safety testing into a multi-tiered system. While organizations like SecureBio handle narrower biosecurity reviews requiring explicit company approval to publish, METR sits at the top tier with full employee-level access and autonomous publishing rights. Reporting by [Ars Technica](https://arstechnica.com/ai/2026/10/anthropic-gives-outside-ai-watchdog-employee-like-access/) highlights that Anthropic is not paying METR for this arrangement, as the research group maintains a strict policy against accepting direct financial donations from AI companies or their staff.

Banking-Style Supervision and Industry Pressures

To justify the arrangement and address potential criticisms regarding transparency and safety, Anthropic compared the auditing process to standard practices in the banking sector. The company likened the structure to how financial regulators assign resident examiners to work directly inside the world's largest banks to monitor systemic risks before they affect the public.

The new oversight comes at a critical juncture for Anthropic as the lab navigates growing industry pressure and prepares for an expected initial public offering valued at over $2 trillion. By voluntarily opening its doors to comprehensive third-party evaluation, Anthropic aims to bridge the transparency gap between internal risk understanding and public accountability.

Sources

Ars TechnicaAnthropic gives outside AI watchdog “employee-like” access
TechCrunchAnthropic is letting METR researchers look under the hood with employee-level access