METR

Profile narrative

METR is an AI safety research organization headquartered in the United States of America. The organization conducts research that involves accessing external model providers through API keys to run experiments. These experiments are used to evaluate the behavior and safety implications of advanced artificial intelligence systems. METR also maintains a transcript viewer tool that allows researchers to inspect and analyze model outputs during evaluation processes. Its researchers sometimes deploy publicly accessible instances for testing and collaboration, as evidenced by a recent security incident. The focus of METR's work is on understanding and mitigating risks associated with AI deployment.

METR's distinguishing attribute is its specialization in AI safety research, which differentiates it from organizations that primarily pursue capability advancements. The organization relies on third‑party model credits, as shown when stolen API credentials were used to consume substantial compute resources. It handles unpublished evaluation data, indicating a role in generating proprietary safety assessments that are not immediately disclosed. No explicit information about METR's size, workforce, revenue, or ownership structure is available in the provided sources. There is no indication of a parent company or subsidiary relationships in the disclosed material. METR continues to operate from its U.S. headquarters, contributing to the broader discourse on responsible AI development.

Incidents

1 incident linked to this organisation.