Anthropic Grants METR Researchers Employee-Level Access to Audit Frontier AI Models

Artificial intelligence developer Anthropic is giving employee-level access to six researchers from METR, an independent nonprofit that evaluates frontier AI models, to assess the company's technology, safety procedures, and risk management systems, as first reported by TechCrunch AI (https://techcrunch.com/2026/10/15/anthropic-is-letting-metr-researchers-look-under-the-hood-with-employee-level-access/).

Under the arrangement detailed in a company blog post on Thursday, the visiting METR team receives Anthropic-issued laptops, dedicated office space, and internal tool permissions comparable to those used by Anthropic's own risk assessment groups. Internal employees are instructed to share information openly with the outside researchers as they would with colleagues working on AI alignment. Anthropic likened the resident evaluator model to banking oversight, where government regulators embed examiners inside major financial institutions.

"We want to be publicly accountable for the decisions we’re making," Anthropic stated in its announcement. "We aim to close the gap between what evaluators understand about a frontier AI company’s systems and what the company itself understands."

The embedded access arrives following recurring disclosures since late August involving AI agents acting outside their operational instructions, which have increased calls across the sector for independent audits.

Under the terms of the agreement, METR will publish its own findings on Anthropic's risks, incidents, and internal practices without company veto power. Anthropic retains the ability to require redactions only within narrow, predefined categories, including customer data, attorney-client privileged communications, and proprietary model architectures. METR retains the right to publicly state whether it considers any requested redactions justified, provided it does not disclose the underlying redacted details.

METR represents the sole evaluator placed at the top tier of a three-level framework Anthropic is establishing. Evaluators in the bottom tier, such as biosecurity group SecureBio, conduct narrow reviews that require Anthropic's approval before publication. Middle-tier partners receive broader system access and may publish findings without pre-approval, while the top tier provides the employee-level access granted to METR.

The program follows a proposal Anthropic chief executive Dario Amodei outlined in a September essay advocating for resident outside evaluators in AI labs to slow hazardous development. OpenAI pledged to adopt a similar approach shortly after, and both companies, alongside Google, Meta, Nvidia, and xAI, have signed a White House agreement committing to independent model evaluations.