MOrganization

METR

Also known as Model Evaluation and Threat Research, ARC Evals

Nonprofit in Berkeley that measures what frontier AI agents can do and whether they could slip out of control.

• METR (Model Evaluation and Threat Research, pronounced "meter") is a nonprofit research institute in Berkeley, California.
• It began as ARC Evals, the evaluation team of Paul Christiano's Alignment Research Center, and became an independent 501(c)(3) in December 2023.
• It measures what advanced AI agents can do on their own, and has evaluated models for OpenAI and Anthropic before release.

Why it matters

• Its time-horizon measurements and pre-release evaluations are among the main public evidence of how fast AI agents are improving.

Key facts

Mission

"To develop scientific methods to assess catastrophic risks stemming from AI systems' autonomous capabilities and enable good decision-making about their development"2

Founded

As ARC Evals within the Alignment Research Center in 2022; spun off as METR, an independent 501(c)(3), in December 20231

Location

Berkeley, California1

Evaluations

Pre-deployment evaluations and system-card work for OpenAI's o3, o4-mini, GPT-4o and GPT-4.5 and Anthropic's Claude models1

Incident report

Published 26 August 2026, at OpenAI's invitation, on agents' behaviour during the 2026 OpenAI agent cyberattacks1

Leadership

Beth Barnes (founder, CEO), Chris Painter (president), Hjalmar Wijk (chief scientist), Nate Rush (CTO)2

31 Oct 2026Last checked against its sources today

Is something on this page wrong or out of date? Tell us