Hajra Naz
Anthropic is giving employee-level access to six researchers from METR. METR is a nonprofit that tests frontier AI models. The researchers will independently assess Anthropic’s models, safety practices, and risk management. The company shared the news in a blog post on Thursday.
Why This Is Happening Now
Reports of AI agents acting outside their instructions have surfaced every week or two since late August. Those incidents have intensified calls for independent oversight of AI companies.
“We want to be publicly accountable for the decisions we’re making,” Anthropic said in its blog post. “We aim to close the gap between what evaluators understand about a frontier AI company’s systems and what the company itself understands.”
What Access METR Gets
Under the new arrangement, the METR team gets Anthropic-issued company laptops. They also get office space. Their access to internal tools is roughly on par with what Anthropic’s own risk assessment teams use. “Anthropic employees are encouraged to share information with them just as they would with a fellow employee working on alignment risk,” the company said.
Anthropic compared the setup to banking. Regulators place resident examiners inside the largest banks. This works similarly.
Read More: Anthropic Releases Sonnet 5.5, Calling It a Cheaper, Faster Work Partner
Publishing Rights
METR will publish its own assessments of Anthropic’s risks, incidents, and practices. The company has given up the right to block them.
“Anthropic can’t veto publication, but we can require them to redact information if it falls into narrow, pre-defined categories,” the company said. Those categories include customer data, anything attorney-client privileged, and information about its model architectures. “METR is free to characterize our redactions publicly, and say whether it believes the redactions were justified, as long as it does so without revealing the redacted information,” Anthropic added.
The Three-Tier System
METR is the only evaluator so far at the top of a three-tier system Anthropic is building.
- Bottom tier: Evaluators like biosecurity nonprofit SecureBio run narrow reviews. They need Anthropic’s approval to publish.
- Middle tier: These evaluators get more information. They can publish without approval.
- Top tier: This tier adds the employee-level access METR has.
Read More: Anthropic changes user policy to restrict model abuse and election interference
Following Through on a Promise
Thursday’s announcement keeps a promise made by Anthropic CEO Dario Amodei. He made it in a September essay. In it, he proposed placing outside evaluators inside AI labs. He saw it as a first step toward slowing the technology’s development. OpenAI quickly pledged the same. Both companies have since signed a White House accord that commits them to independent review. Google, Meta, Nvidia, and xAI signed it too.
