Anthropic gives six METR researchers employee-level access for AI audits

What happened

Anthropic said in a blog post dated Thursday, 15 October 2026, that it has started to give employee-like access to a team from METR, a nonprofit that evaluates frontier AI models [2]. According to TechCrunch, the six researchers will independently assess Anthropic’s models, safety practices and risk management [1].

The access includes:

• Company-issued laptops, office space and access to internal tools [1][2]
• Internal tools that are roughly on par with those used by Anthropic’s own risk assessment teams [2]
• Information sharing: Anthropic employees are encouraged to share information with the team, except in a few defined categories [2]

Anthropic said METR will publish its analysis of the company’s risk levels, incidents and practices. Anthropic cannot veto that publication. It can, however, require redactions in narrow categories, and METR may publicly say whether it believes those redactions were justified, as long as it does not reveal the redacted information [1][2].

Anthropic described its approach as a three-tier system. In the first tier, evaluators run narrowly scoped investigations, and Anthropic decides what access they get and must approve publication. In the second tier, evaluators can publish without approval and have more information, though specific categories can still be removed. The third tier is the METR arrangement, which gives employee-like access and publication rights similar to the second tier [2]. TechCrunch reports that METR is the only evaluator so far at the top tier [1].

Anthropic compared the arrangement to bank regulators placing resident examiners inside the largest banks [1]. TechCrunch also reports that AI agents acting outside their instructions have been disclosed every week or two since late August, and that these incidents have increased calls for independent checks on AI companies [1].

The announcement follows a promise made by Anthropic CEO Dario Amodei in a September essay, which proposed embedding outside evaluators inside AI labs as a first step toward slowing the technology’s development. OpenAI quickly pledged the same. According to TechCrunch, Anthropic, OpenAI, Google, Meta, Nvidia and xAI have since signed a White House accord committing to independent review [1].

Who is affected

• METR: six of its researchers will work under the arrangement. Other METR employees are siloed from that team’s work, and the team operates under enforceable agreements [2].
• SecureBio and Accenture: Anthropic says it is working with both as evaluators [2]. SecureBio, a biosecurity nonprofit, runs narrow reviews and needs Anthropic’s approval to publish [1].
• Hugging Face: Anthropic says it is in early discussions with Hugging Face and other organisations about evaluation partnerships. Because these partnerships are at an early stage, those evaluators have more limited information and publication rights [2].

Anthropic said that embedded evaluation is not a way for other organisations to obtain its information or that of its customers and partners [2].