Anthropic Grants METR Deep Access for Independent AI Safety Evaluations
Anthropic has announced a significant transparency initiative by granting METR, an organization focused on evaluating AI systems, employee-level access to its internal operations. Under the agreement, METR researchers will be able to conduct their own assessments of Anthropic's safety practices, risk management procedures, and any incidents that occur with the company's AI models.
The arrangement is notable for the degree of independence it affords METR. Anthropic has surrendered its right to block or suppress publication of any findings, meaning the research organization can release its assessments publicly regardless of whether the results are favorable to the company. This puts METR in a position to serve as an external auditor with real insight into frontier AI development.
METR has built its reputation on developing evaluation frameworks designed to assess the capabilities and potential risks of advanced AI systems. By gaining direct access comparable to that of Anthropic's own staff, the organization can conduct more thorough and nuanced evaluations than would be possible through external observation alone.
The move reflects an ongoing industry conversation about how to balance AI development speed with safety accountability. Independent evaluation mechanisms like this one aim to provide stakeholders—including regulators, researchers, and the public—with credible, third-party assessments of how leading AI companies are managing risks as systems grow more capable.