AI company Anthropic has entered into a transparency agreement with METR, an independent research organization focused on evaluating the safety and capabilities of frontier AI models. This partnership provides METR researchers with a level of access typically reserved for internal employees, allowing them to examine the company's models, safety protocols, and operational practices directly.

A significant component of this arrangement is the removal of editorial control. Anthropic has formally relinquished its right to review or block the reports produced by METR. This means METR is permitted to publish its own assessments regarding Anthropic’s risk management, potential incidents, and general safety practices. The move is intended to provide the public and regulators with an objective view of how the company manages the risks associated with advanced artificial intelligence.

While the industry has faced ongoing pressure to demonstrate safety, this collaboration represents a shift toward third-party oversight. By allowing an external entity to operate with 'employee-level' access and granting them the autonomy to publish findings independently, Anthropic aims to establish a higher standard for accountability in the development of frontier AI models. METR’s upcoming assessments are expected to cover a range of safety metrics, providing a clearer picture of the internal safeguards currently in place at the company.