Senator Welch Introduces Bipartisan TRAIN Act for AI Transparency
Senator Peter Welch has introduced the Transparency and Responsibility for Artificial Intelligence Networks Act (TRAIN Act) on July 24, 2025. This bipartisan legislation aims to bring clarity to the use of copyrighted works in training artificial intelligence models. The bill has received support from Republican Senators Marsha Blackburn and Josh Hawley, as well as Democratic Senator Adam Schiff.
The TRAIN Act targets generative AI models that create synthetic content like images, videos, audio, and text. These models often use copyrighted works for training, but the developers—such as OpenAI, Google DeepMind, Anthropic, and Meta—generally keep their training data private due to proprietary or competitive reasons. The legislation addresses this lack of transparency by establishing a process for copyright holders to request subpoenas compelling AI developers to disclose their training materials.
The TRAIN Act defines AI models, developers, and training materials, and sets standards for substantial modifications. It introduces enforcement mechanisms, including rebuttable presumptions against non-compliant developers. The bill incorporates Federal Rules of Civil Procedure for subpoena issuance, service, and enforcement, while maintaining expeditious disclosure requirements. Procedurally, copyright owners must demonstrate good faith belief and file sworn declarations to request subpoenas. The legislation also includes confidentiality provisions and sanctions for bad faith requests.
The TRAIN Act's introduction comes amidst high-profile copyright lawsuits challenging AI companies' training practices. By targeting generative AI models and establishing a clear process for copyright holders to seek disclosure, the legislation aims to promote transparency and responsibility in the use of copyrighted works for AI training.
