METR

METR (Model Evaluation and Threat Research) is a US nonprofit based in Berkeley, California, that develops scientific methods to assess the catastrophic risks stemming from AI systems' autonomous capabilities. It became an independent organization in December 2023 after spinning out of the Alignment Research Center (having previously operated as the ARC Evals team). METR runs evaluations of frontier AI models from leading labs (including OpenAI, Anthropic, and Google DeepMind), studies AI agent autonomy, and develops the 'time horizon' benchmark measuring the length of tasks models can complete autonomously. The organization also contributes to risk-governance frameworks such as Responsible Scaling Policies.