Anthropic gives METR insider access to test its AI; OpenAI pledges to follow

AI company Anthropic has given six researchers from METR, a non-profit that tests advanced AI models, employee-level access to its systems so they can assess its models and safety practices from the inside. Anthropic announced the arrangement in a blog post on Thursday, October 15, 2026. According to the company, the METR staff have been given company hardware, office space and access to internal tools roughly on par with what Anthropic's own risk-assessment teams use. METR can publish its findings without Anthropic's approval; Anthropic can ask for redactions only in narrow, pre-defined categories such as customer data, legally privileged information and details of model architecture. OpenAI has said it will follow suit, and reports say Anthropic, OpenAI, Google, Meta, Nvidia and xAI have signed a White House accord committing to independent review of their models.

Outside groups usually see only what AI companies choose to publish about their own testing. Giving an independent group deep, ongoing access, and the right to publish what it finds, is a step towards closer outside scrutiny. If it works, it could make claims about AI safety easier to check.

The arrangement also sets a marker for the rest of the industry. OpenAI's pledge to follow, and the White House accord signed by six major companies, suggest that embedded outside evaluators could become a norm for frontier AI developers in the United States.

There are limits, and critics have pointed them out. Anthropic chose the evaluator and controls the scope of access, and it can still require some redactions. Commentators note that the accord commits companies to outside evaluators but does not set rules for how those evaluators must work. And as of the coverage so far, METR has not yet published a report under these terms, so the real test will be what it says and how freely it can say it.

For Indian policymakers and users, the model offers one possible approach to checking the claims of global AI companies.

Short-term: METR will begin assessing Anthropic's models and safety processes; its first public report will show how much the arrangement reveals.

Long-term: If embedded independent review spreads across the industry, it could become a standard part of how frontier AI models are checked before and after release, and could inform regulators in other countries.

Who is affected: AI developers, independent evaluators and researchers, regulators in the United States and abroad, and the businesses and people who use AI systems.

Questions and Answers