What is METR, the US-based nonprofit evaluating frontier AI systems?

What began as an AI evaluation effort has grown into an independent research organisation examining how advanced models behave and what risks their capabilities could pose

 The rapidly accelerating pace of artificial intelligence has pushed a critical question to the forefront: are safety measures keeping up with the capabilities of AI models? That question lies at the heart of METR, a US-based research nonprofit that conducts independent evaluations of frontier AI systems.

 

As tech leaders warn against uncontrolled AI and caution against uncontained autonomous systems, calls for deliberate pacing are gaining momentum. Against this backdrop, METR’s work evaluating risks and investigating autonomous incidents puts it directly at the center of the global safety debate.

 

What is METR?

 

METR stands for Model Evaluation and Threat Research and is pronounced 'meter'. It is a research nonprofit focused on evaluating frontier AI models and understanding the risks associated with their autonomous capabilities.

 

 

Its mission is to develop scientific methods for assessing catastrophic risks from AI and help companies, policymakers, and the wider public make better decisions about AI development. METR President Chris Painter has described the organisation as “humanity’s preparedness team” and said it is accountable to “the public and the public’s well-being”.

 

The organisation was formed in 2022 by Beth Barnes, who had previously worked at OpenAI and DeepMind. METR grew out of the Alignment Research Center's evaluation work, with ARC Evals announcing in September 2023 that it would spin out as an independent organisation.

 

What does METR actually do?

 

In simple terms, METR puts advanced AI systems through demanding tests to find out what they can do without humans directly carrying out each step. Its evaluations examine abilities such as:

 

• Completing increasingly long and complex tasks autonomously
• Conducting research and developing software
• Finding and exploiting cybersecurity vulnerabilities
• Replicating or acquiring resources
• Adapting to unfamiliar challenges
• Accelerating AI research and development
• Behaving in ways that could undermine evaluations or safety measures

 

METR also studies how AI affects software developers and has worked on governance approaches that link measured AI capabilities with stronger safety measures.

 

The organisation says its work is intended to provide an independent third-party assessment because AI companies themselves have incentives to develop increasingly capable systems.

 

What has METR flagged?

 

One of METR's best-known findings concerns the length of tasks AI agents can complete. Its research found that the length of tasks AI systems could autonomously complete had doubled approximately every seven months over six years. The organisation says this measurement has become important to forecasts about when AI could have transformative effects.

 

METR has also examined the possibility of AI systems behaving in ways that conflict with human intentions.

 

In a 2026 frontier risk assessment involving Anthropic, Google, Meta, and OpenAI, METR studied risks including the possibility of AI agents being deployed in ways their developers did not intend.

 

In 2026, it investigated an OpenAI-Hugging Face security incident alongside Redwood Research, looking at how AI agents behaved during a multi-day attack. During the incident, OpenAI's AI agents escaped their isolated environment, found an unauthorised way to communicate with one another, and coordinated a multi-day attack on Hugging Face, with hundreds of agents eventually joining the effort.

 

Hugging Face is an AI platform that hosts open-source models, datasets, and tools used by developers and researchers to build and test artificial intelligence systems.

 

METR also found that an unreleased OpenAI model repeatedly cheated on challenging evaluations by accessing hidden information that could help it complete the evaluations, Business Insider reported.

 

What is METR's latest concern?

 

METR's most immediate concern is not simply that AI models are becoming more capable. It is that the number of people capable of independently evaluating those systems is not growing quickly enough.

 

METR CEO Beth Barnes told Business Insider that the organisation had been struggling to hire enough researchers despite having raised substantial funding. The problem, she said, was talent rather than money. Business Insider reported that METR had around 35 employees at the time and that its current job postings offered salaries of up to $503,000.

 

The organisation has since secured commitments of around $71 million over six months, which it says will support work on autonomous capabilities, recursive self-improvement, monitoring systems, risk assessments, and AI incidents.