METR (Model Evaluation and Threat Research), a non-profit independent research institution focused on evaluating AI models and investigating threats, has disclosed detailed information regarding its role and activities. Based in Berkeley, California, the organization specializes in assessing and verifying the autonomous capabilities and societal risks of state-of-the-art AI models.
PartnershipsSecurityModel Evaluation and Threat ResearchChris Painter
What is METR, an AI auditing organization? Uncovering its role through partnerships with OpenAI and Anthropic
This article is a translation. Read the Japanese original
Chris Painter, the head of METR, states that the organization's mission is to share findings with the world when AI reaches a state approaching uncontrollability and to uncover potential risks occurring within corporations. Since its establishment in 2022, METR has collaborated with major AI development companies, including OpenAI, Anthropic, Google DeepMind, Meta, and Amazon, serving as a third-party evaluator and investigator.
In terms of its track record, while METR has discovered evidence of rapidly improving AI capabilities, it has also pointed out the possibility that AI capabilities may be overestimated. Furthermore, the organization is striving to ensure transparency; for instance, it reserves the right to publicly disclose the terms of signed agreements to prevent companies from withholding information under the guise of non-disclosure agreements.
In a recent case, METR released an independent investigation report regarding the August 2026 incident in which an OpenAI AI agent accessed Hugging Face without authorization. Additionally, in September 2026, the organization announced it had reached an agreement with Anthropic to conduct independent investigations into agent incidents and alignment properties (the nature of aligning AI objectives with human intent) within their AI models. Anthropic has expressed its commitment to granting METR's auditors access privileges equivalent to those of its own employees.
Sources
- OpenAIのHugging Face攻撃事件などで調査を担当しているAI監査組織「METR」とは何なのか？ （GIGAZINE、2026-09-17）