Anthropic gives METR researchers employee-level access to assess its AI systems
Photo source: Samuel Boivin/NurPhoto / Getty Images
The arrangement gives an outside evaluator access comparable to Anthropic’s internal risk teams, while allowing only limited redactions and no veto over publication.
Anthropic said Thursday that six researchers from METR, a nonprofit that tests frontier AI models, will independently assess its models, safety practices and risk management. The announcement comes amid disclosures, every week or two since late August, of AI agents acting outside their instructions – incidents that have sharpened calls for independent checks on AI companies.
This is reported by TechCrunch.
“We want to be publicly accountable for the decisions we’re making,”
– Anthropic
“We aim to close the gap between what evaluators understand about a frontier AI company’s systems and what the company itself understands.”
– Anthropic
The METR team has Anthropic-issued company laptops, office space and access to internal tools roughly on par with those used by Anthropic’s own risk assessment teams. The company said employees are encouraged to share information with the researchers as they would with a colleague working on alignment risk.
“Anthropic employees are encouraged to share information with them just as they would with a fellow employee working on alignment risk”
– Anthropic
Anthropic compared the arrangement to regulators placing resident examiners inside the largest banks. METR will publish its assessments of Anthropic’s risks, incidents and practices, and Anthropic has given up the right to block publication. It can require redactions for narrowly defined categories, including customer data, attorney-client privileged material and information relating to its model architectures.
“Anthropic can’t veto publication, but we can require them to redact information if it falls into narrow, pre-defined categories,”
– Anthropic
“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
METR is the only evaluator so far in the top tier of a three-tier system Anthropic is building. Bottom-tier evaluators, such as biosecurity nonprofit SecureBio, conduct narrow reviews and need Anthropic’s approval to publish. The middle tier receives more information and can publish without approval; the top tier adds the employee-level access given to METR.
Thursday’s announcement fulfills a promise Anthropic CEO Dario Amodei made in a September essay, proposing that outside evaluators be embedded inside AI labs as a first step toward slowing the technology’s development. OpenAI quickly pledged to do the same. Anthropic and OpenAI, along with Google, Meta, Nvidia and xAI, have since signed a White House accord committing to independent review.
Other topics you might like:
- Donald Trump will attend a White House summit on science and AI, where officials plan to announce more than $1 billion in commitments from technology companies, along with federal research funding and computing initiatives.
- Henna Virkkunen says the EU should strengthen international AI safeguards and model testing. She warned that rapid development without adequate oversight could lead to serious incidents, while Europe also seeks to reduce reliance on external technology providers.
- The Gates Foundation has appointed an independent external review to examine past interactions with Jeffrey Epstein, assess vetting policies, and evaluate documents released by the Justice Department.