Anthropic Grants METR Inside Access to Audit AI Models and Safety Practices
Anthropic has given the nonprofit METR direct access to review its AI models and safety practices, the company announced, in a move that expands independent scrutiny of frontier AI development.
Anthropic has granted a team from the nonprofit research organization METR inside access to review its artificial intelligence models and safety practices, the company announced. The arrangement gives external researchers a direct view of how one of the world's leading AI developers builds, tests, and governs its most capable systems.
The decision marks a notable step in the growing effort to subject frontier AI development to independent examination. METR, which specializes in evaluating advanced AI systems, will be able to inspect Anthropic's models and the internal safety procedures that surround them, rather than relying solely on public statements or limited external testing.
Anthropic has positioned itself as a developer focused on AI safety, and the access granted to METR extends that posture into practice. Independent auditors can probe how models behave, what safeguards are in place, and whether the company's stated safety commitments hold up under scrutiny. Such reviews are increasingly seen as important because the most powerful AI systems are developed behind closed doors, leaving outside observers with little ability to verify claims about their capabilities or risks.
The announcement comes amid a broader debate over how AI companies should be held accountable. Governments, researchers, and civil society groups have called for greater transparency, standardized evaluations, and third-party oversight as AI systems become more capable and more widely deployed. METR's role as an independent evaluator places it at the center of that debate, and its findings could inform how other labs approach external review.
For Anthropic, opening its models and safety practices to an outside team carries both risks and benefits. It exposes internal processes to critical examination, but it also builds credibility at a time when public trust in AI developers is fragile. The company has argued that rigorous, independent evaluation is necessary for the responsible development of advanced AI, and the METR arrangement is a concrete expression of that argument.
The scope of the access and the specific models involved were not detailed in the announcement. It remains unclear how long the review will last, what METR will publish, and whether the findings will be made public in full. Those questions matter because the value of any audit depends on how independent it is, how deep it goes, and how transparently the results are shared.
METR's work has focused on measuring the capabilities and risks of frontier AI systems, including whether models can carry out complex tasks autonomously or resist human oversight. Access to a major developer's models and internal safety practices gives the organization a rare vantage point. Independent evaluators typically work with limited interfaces, so deeper access can reveal behavior that standard testing misses.
The move also reflects a wider shift in the AI industry, where voluntary commitments and external partnerships are emerging alongside formal regulation. Policymakers in Europe, the United States, and elsewhere are drafting rules that would require safety testing and transparency from advanced AI developers. Private arrangements like the one between Anthropic and METR may serve as models, or as stopgaps, while those rules take shape.
For readers concerned with the moral and human questions raised by technology, the arrangement touches on a central issue: who gets to decide whether a powerful new technology is safe, and on what evidence. Anthropic's decision to let an outside nonprofit look inside its operations is an attempt to answer that question with verification rather than assertion. Whether it becomes a durable standard or a one-off gesture will depend on what METR finds and what Anthropic does with the results.
4
