Anthropic is letting METR researchers look under the hood with employee-level access

Anthropic has granted the Model Evaluation and Threat Research (METR) organization deep, internal access to its systems, effectively relinquishing its right to censor or block independent findings. This move represents a significant push for transparency in AI safety, as outside researchers can now audit Anthropic’s operational risks and internal practices as if they were staff members. By opening its doors to external accountability, the company is attempting to establish a new gold standard for safety audits in the frontier model industry.