METR

Profile narrative

METR, also known by its abbreviation METR, is an AI safety research organization whose work centers on evaluating and assessing the capabilities, risks, and behaviors of advanced artificial intelligence systems. The organization's publicly documented activity includes operating research infrastructure and tooling, including publicly accessible instances used by the AI safety community, as well as a public transcript viewer associated with evaluation work. This combination of operational tooling and research output suggests that METR occupies a hands-on role in AI safety rather than functioning solely as an advisory or policy body. Its core focus on AI safety situates it within the broader ecosystem of organizations working to understand and measure the implications of increasingly capable AI models.

The most prominent documented incident involving METR occurred on March 1, 2026, when an attacker exploited a fail-open flaw in a researcher's publicly accessible instance to steal an API key belonging to the organization. Over a period of roughly three weeks, the threat actor used the stolen credentials to consume approximately $600,000 worth of model credits supplied by an unnamed provider. A separate flaw was identified in a public transcript viewer, which could have exposed unpublished evaluation data; however, METR found no evidence that sensitive information was actually accessed in either incident. This incident demonstrates both the organization's reliance on third-party compute providers for its research and the risks associated with publicly exposed research infrastructure. The substantial financial scale of the credit consumption, amounting to hundreds of thousands of dollars, indicates that METR's research activities involve significant computational resources.

As a research-focused entity operating in the AI safety sector, METR distinguishes itself through its direct engagement with model evaluation and infrastructure. The available material does not specify ownership structure, parent organizations, subsidiaries, funding sources, headcount, or geographic footprint, so any conclusions regarding its scale or governance cannot be drawn from the provided context. Likewise, no information is available about its regulatory role, formal accreditation, or specific market positioning beyond its identification as an AI safety research organization. The organization appears to maintain a public-facing presence through its instances and transcript viewer, which is consistent with a research culture that emphasizes transparency and community access to its tools and findings.

Incidents

1 incident linked to this organisation.