National Institute of Standards and Technology

A physical science laboratory inside Commerce

NIST, the National Institute of Standards and Technology, is a physical science laboratory that now forms part of the U.S. Department of Commerce ^[1]. It describes itself as one of the nation's oldest physical science laboratories ^[1]. Its stated mission is to promote U.S. innovation and industrial competitiveness by advancing measurement science, standards and technology ^[1], and it lists three core competencies: measurement science, rigorous traceability, and the development and use of standards ^[1].

Those three terms describe a fairly specific job. Measurement science is the study of how to measure something reliably — which instrument, which procedure, how much uncertainty to attach to the answer. Traceability is what makes a measurement checkable: a value is traceable when it can be connected, step by step, back to a defined reference, so that two laboratories measuring the same thing can tell whether their results agree. Standards are the written agreements — definitions, procedures, reference materials — that allow those comparisons to happen between organisations rather than inside a single one. A laboratory that does all three acts as the common reference point that other laboratories calibrate against.

Why Congress created it

Congress established the agency in 1901 ^[1]. The motivation was economic rather than scientific. The problem it was meant to address was a second-rate measurement infrastructure that lagged behind the United Kingdom, Germany and other economic rivals, and that lag was treated as a challenge to U.S. industrial competitiveness ^[1].

The reasoning behind that kind of concern is not hard to reconstruct. An industrial economy runs on parts and materials that have to fit together and be bought and sold by specification. If one company's unit of length, mass or electrical resistance is not quite the same as another's, then assemblies fail, contracts become disputes, and testing results cannot be compared across borders. A national laboratory whose job is to hold the definitions and provide the reference measurements gives everyone else something to agree on, without each firm having to build its own metrology infrastructure. That is a public good in the strict sense: cheap to share, expensive to duplicate, and useful to competitors and customers alike.

From calibration to AI safety

The U.S. AI Safety Institute was located within the Department of Commerce at NIST, and was established following the 2023 executive order on AI ^[2]. NIST said the institute signed agreements with Anthropic and OpenAI that give it access to major new models before and after public release ^[2]. NIST records that in June 2025 the AI Safety Institute was re-established as the Center for AI Standards and Innovation by Secretary of Commerce Howard Lutnick ^[2][3].

The AI Safety Institute also announced the Testing Risks of AI for National Security (TRAINS) Taskforce, which it chairs. Initial members included the Department of Defense, the Department of Energy and ten of its national laboratories, the Department of Homeland Security, and the National Institutes of Health ^[3]. The taskforce was set up to develop new AI evaluation methods and benchmarks, and to conduct joint national security risk assessments and red-teaming exercises ^[3]. Red-teaming, in general use, means deliberately attacking or probing a system to find the ways it fails before someone else does.

Where it sits

The move into AI evaluation is a recognisable extension of the older role. Evaluating a model is a measurement problem: you need a procedure, a reference point and a way for different labs to get comparable answers, which is what standards bodies exist to supply. The difference is that the objects being measured are new and changing quickly, and the institutions taking part — defence, energy, homeland security, health — are the ones that would have to act on the results. NIST's own description of itself still centres on measurement science, traceability and standards ^[1]; the AI work is what those competencies look like when the technology under discussion is a trained model rather than a physical instrument.