Anthropic Grants METR Full Internal Access for AI Safety

Anthropic granted a team of six employees from the nonprofit AI evaluator METR employee-level access to its internal systems, hardware, and office space. This arrangement allows METR to independently assess the company's advanced models, risk management, and safety practices using tools comparable to those used by internal risk assessment teams.

Under the agreement, METR may publish independent assessments of risks and incidents without prior approval from Anthropic. The company retains the right to request limited redactions of customer data, privileged information, and specific model architecture details.

The initiative addresses industry concerns that AI laboratories maintain excessive control over independent reviews through restrictive publication policies and limited access to sensitive data.