READ AI Models Act (H.R.6461, 2025): A "Nutrition Label" for AI Models

Artificial intelligence is spreading across schools, governments, and small businesses, yet users lack a common yardstick to confirm what a given model was trained on and how thoroughly it was tested. Large firms frequently publish detailed "model cards," while smaller developers and adopters may lack the capacity to produce or interpret them. That information asymmetry makes judgments about safety and fitness-for-purpose harder.

H.R.6461, the READ AI Models Act, is a bipartisan bill that responds to this with standardization. At its core is a directive to the National Institute of Standards and Technology (NIST), the federal agency for measurement and technical standards, known for voluntary guidance such as its AI Risk Management Framework. As a pilot program, NIST would create a modular documentation template for describing AI models. Captured fields would include model name, developer, release date, training-data cutoff, supported languages, and terms of service, structured so users can apply only the parts relevant to their use case and scale. NIST would also develop technical guidance pointing to evaluation metrics and benchmarks appropriate to different model types, drawing on voluntary consensus standards (self-regulatory standards developed by stakeholder agreement) and industry best practices.

The bill also specifies the process. NIST would work with private companies, universities, nonprofits, and federal agencies, and publish drafts in the Federal Register (the government's official journal of notices) for a public-comment period of not less than 60 days. Within 12 months of enactment, NIST would submit a report to Congress assessing the pilot's effectiveness and recommending whether to make it permanent. Rather than imposing mandates, the design is modest: create a usable common format first and test its adoption.

As for current status, the bill was introduced in the House on December 4, 2025, and referred to the House Committee on Science, Space, and Technology. Committee review, a floor vote, Senate action, and a presidential signature all remain ahead, so whether it becomes law depends on further proceedings. Still, its approach is notable: it advances AI transparency not through compulsion but through a practical common documentation format, which could become a foundation for procurement and accountability for organizations that use AI.