METR

METR (pronounced 'meter') is a nonprofit research institute based in Berkeley, California, that develops scientific methods to assess whether and when AI systems might threaten catastrophic harm to society. Founded with the mission to advance the science of AI evaluations, METR works to empirically measure the autonomous capabilities of frontier AI systems and evaluate the risks they pose to global stability.

The organization conducts rigorous evaluations assessing the extent to which AI systems can autonomously carry out substantial tasks, including general-purpose capabilities like conducting research or developing applications, as well as concerning capabilities such as executing cyberattacks or making themselves difficult to shut down. METR's research on 'time horizons' - measuring how long of a human task AI agents can complete - has become central to forecasts of when AI will have transformative impacts on society.

METR partners with leading AI companies including OpenAI, Anthropic, Google DeepMind, Meta, and Amazon to pilot frontier risk assessments. The organization also contributes to governance standards, having prototyped the Responsible Scaling Policies approach that has since been adopted by nine leading AI developers. As an independent third-party, METR publishes its research and risk assessments openly, believing that transparency and empirical rigor are essential for helping policymakers and civil society understand the risks from advanced AI systems.