HR-7209-119
Referred to the House Committee on the Judiciary.
Sponsored by Madeleine Dean (D-PA)
What it does
The TRAIN Act would add a new section to federal copyright law (Title 17) allowing copyright owners to request a court clerk to issue an administrative subpoena to AI developers. The subpoena would compel developers to disclose copies of, or records identifying, which of the requester's copyrighted works were used to train a generative AI model. The requester must submit a sworn declaration of good-faith belief that their works were used; developers who fail to comply face a rebuttable presumption that they copied the work; and bad-faith subpoena requests may result in court-imposed sanctions.
Who benefits
Copyright holders — including individual authors, musicians, visual artists, photographers, journalists, and screenwriters — who want to determine whether their works were used in AI training without their consent. Large media companies, publishers, and record labels that hold large catalogs of copyrighted content. Licensing agencies and collective rights organizations. Attorneys specializing in copyright litigation, who would gain a new discovery tool. Potentially, creators who could use disclosed records as the evidentiary foundation for infringement lawsuits.
Who is hurt
AI developers and companies — including large technology firms and startups — that would face compliance costs, legal exposure, and potential disclosure of proprietary training datasets. Open-source AI projects and academic researchers who develop generative models could face subpoena burdens even for non-commercial work (though noncommercial end users are explicitly excluded). Businesses that rely on AI-generated content and whose tools could face legal uncertainty. Consumers who may see slower AI development or higher costs passed through from compliance. Third-party dataset curators who fall within the bill's definition of "developer."
Supporters argue
Supporters argue that AI developers have trained models on vast quantities of copyrighted material — books, images, music, and journalism — without licensing or compensating creators, and that copyright owners currently have no practical way to discover whether their specific works were used. They contend the bill fills a critical evidentiary gap: without access to training records, creators cannot even determine whether they have a viable infringement claim, let alone pursue one. The subpoena mechanism mirrors existing tools in copyright law (such as the DMCA § 512(h) subpoena) and includes meaningful safeguards against abuse, including a sworn good-faith declaration requirement and sanctions for bad-faith requests.
Opponents argue
Opponents argue that the bill's "subjective good faith belief" standard for triggering a subpoena is too low, effectively allowing any copyright holder to compel disclosure of a developer's proprietary training data with minimal threshold showing. They contend that training datasets often constitute trade secrets and core intellectual property, and that forced disclosure — even with confidentiality duties — creates serious risks of competitive harm and chilling effects on AI research and development. Critics also argue the rebuttable presumption of copying upon non-compliance is a severe penalty that could coerce settlements even where no actual infringement occurred, undermining due process protections under the Fifth Amendment.
Constitutional context
Congress has broad authority to legislate copyright law under the Intellectual Property Clause (Art. I, §8, cl. 8) and the Commerce Clause (Art. I, §8, cl. 3), as AI training and deployment is plainly economic activity with interstate dimensions under Wickard v. Filburn (1942). The bill's compelled disclosure of training data could raise Fifth Amendment Due Process concerns, particularly around the rebuttable presumption of copying upon non-compliance. Post-Loper Bright (2024), any agency interpretations of the bill's scope would face independent judicial review rather than deference.
Checks and balances
Copyright owners gain a new administrative subpoena tool routed through the judicial branch (district court clerks); AI developers retain the ability to challenge subpoenas under Federal Rules of Civil Procedure, and courts retain authority to impose sanctions on bad-faith requesters and to adjudicate compliance disputes.
Historical precedent
The DMCA § 512(h) subpoena process (1998) established a similar administrative subpoena mechanism allowing copyright owners to compel online service providers to identify alleged infringers, though courts later limited its scope in cases such as Recording Industry Ass'n of America v. Verizon (D.C. Cir. 2003).