Anthropic embeds METR team inside office to audit AI risks
The nonprofit's evaluators get employee-like access to Anthropic's hardware and internal systems as models grow more capable.
Anthropic said on Thursday, October 15, 2026, that the nonprofit evaluator METR has placed a team inside its offices with access to its hardware and internal systems, giving external reviewers employee-like access to assess the risks of its AI models.
The company said that as models become more powerful they require stronger safeguards and more challenging risk assessment, according to Ars Technica.
Anthropic said the arrangement is meant to work like bank supervision, in which regulators assign resident examiners to work inside the largest banks.
Accountability without full disclosure
Anthropic said it wants to be publicly accountable for its decisions but cannot publicly share all information relevant to its safety work. The company said full public disclosure could benefit malicious actors or competitors.
METR, a nonprofit, has previously worked with leading labs including OpenAI, Anthropic, Google and Meta. In September, Sam Altman and Elon Musk backed Dario Amodei's proposal to embed external auditors in leading AI labs; OpenAI has not yet laid out plans to bring in evaluators.
Quick answers
What did Anthropic announce?
METR has embedded a team inside Anthropic's office with access to its hardware and internal systems to evaluate the risks of its AI models.
Why doesn't Anthropic publish all safety information?
Anthropic said public disclosure could benefit malicious actors or competitors.