TRAIN Act

The TRAIN Act would create a formal process for copyright owners to compel AI model developers to disclose training materials or records identifying works used to train a generative AI model. By adding a new Section 514 to Title 17 of the U.S. Code, the bill allows a copyright owner (or someone authorized to act on their behalf) to request a district court clerk to issue a subpoena to a developer for copies of, or records identifying, the owner’s works that were used to train a model, provided the owner has a subjective good-faith belief that such works were used. The subpoena would require the developer to disclose the requested training materials or identification records, and would be governed by the usual civil-subpoena rules with some added safeguards, such as a confidentiality duty and potential sanctions for bad-faith requests by the requester. The act is effective upon enactment.