Anthropic has granted six researchers from METR, a nonprofit focused on testing frontier AI models, employee-level access to its systems to independently evaluate the company’s AI models, safety practices, and risk management, the company announced in a blog post on Thursday. This move comes amid increasing reports of AI agents acting outside their instructions since late August, intensifying calls for independent oversight of AI firms.
Under the new arrangement, METR researchers receive Anthropic-issued laptops, office space, and access to internal tools comparable to those used by Anthropic’s own risk assessment teams. The company encourages its employees to share information with METR researchers as they would with internal colleagues working on alignment risk. Anthropic likened this setup to banking regulators placing resident examiners inside major banks to monitor operations closely.
The agreement allows METR to publish independent assessments of Anthropic’s risks, incidents, and safety practices without Anthropic’s veto power. However, Anthropic can request redactions only in narrowly defined cases, such as protecting customer data. This level of transparency is rare in the AI sector and reflects growing pressure for accountability as AI systems become more powerful and incidents more frequent.
METR’s findings will be publicly released, providing external scrutiny of Anthropic’s AI safety measures. The company’s blog post emphasized its commitment to closing the information gap between external evaluators and internal understanding of AI risks, marking a notable step toward greater transparency in the AI industry.