TRAIN Act
- Sponsor
- Rep. Dean, Madeleine [D-PA-4]
- Committees
- Judiciary Committee (primary)
- Last action
- Jan 22, 2026
Bottom line
The TRAIN Act aims to provide copyright holders with a new tool to ascertain whether their works were used in AI model training, potentially streamlining the discovery phase for future copyright infringement claims.
What it actually does
This bill establishes a new administrative subpoena process under copyright law, allowing copyright owners to compel developers of generative artificial intelligence models to disclose which of the owner's copyrighted works were used in training. The process requires a subjective good faith belief from the copyright owner and mandates expeditious disclosure from developers, with non-compliance leading to a rebuttable presumption of copying.
Proponents argue
Proponents argue this bill provides much-needed transparency for copyright owners who currently lack a clear mechanism to determine if their intellectual property has been used to train AI models. They contend it will help protect creators' rights in the evolving landscape of AI, enabling them to make informed decisions about potential infringement claims without resorting to lengthy and costly litigation for basic discovery.
Opponents contend
Opponents might argue that this subpoena process could be overly burdensome for AI developers, potentially leading to a flood of requests based on subjective good faith beliefs, and could expose proprietary training data or methodologies. They may also raise concerns about the potential for abuse of the subpoena power, even with bad faith sanctions, and the practical challenges of definitively identifying specific copyrighted works within vast training datasets.
The bill is short and relatively straightforward, allowing a reader to quickly grasp its core mechanism and implications.
Section 2, adding new 17 U.S.C. 514(b)-(f)
Establishment of Subpoena for AI Training Data
This provision creates a new legal mechanism allowing copyright owners (or their representatives) to request a subpoena from a U.S. district court clerk. This subpoena would compel developers of generative AI models to disclose copies or records identifying the owner's copyrighted works used in training. The request requires a sworn declaration of a subjective good faith belief that the works were used, and the disclosure is limited to the requester's own works. Clerks are mandated to issue the subpoena if the request is in proper form.
Supporters argue
Supporters argue this process is essential for creators to understand how their work is being utilized by AI systems, providing a necessary first step for protecting their intellectual property rights in the digital age. It offers a more efficient and less costly alternative to full litigation for initial discovery.
Critics contend
Critics may contend that this process could be abused, leading to excessive and burdensome requests for AI developers, potentially revealing trade secrets related to model training and data curation. They might also argue that the 'subjective good faith belief' standard is too low, inviting speculative requests.
Tradeoffs
Balances the need for copyright transparency and protection against the potential burden on AI developers and the protection of proprietary AI development methods.
Section 2, adding new 17 U.S.C. 514(i)
Legal Presumption for Failure to Comply
This provision states that if an AI developer fails to comply with a subpoena issued under this Act, it will create a rebuttable presumption that the developer made copies of the copyrighted work in question. This shifts the burden of proof, requiring the developer to demonstrate they did not copy the work if they fail to provide the requested information.
Section 2, adding new 17 U.S.C. 514(j)
Penalties for Abusive Subpoena Requests
This provision allows a court to impose sanctions on a copyright owner or authorized person if they request a subpoena under this Act in bad faith. It explicitly states that Rule 11(c) of the Federal Rules of Civil Procedure, which governs sanctions for improper filings, shall apply. This aims to deter frivolous or harassing subpoena requests.
The definition of 'developer' includes persons or state/local government agencies that design, code, produce, own, or substantially modify a generative AI model and are involved in training dataset curation or use. 'Substantially modify' is defined as materially changing functionality or performance, including by retraining or fine-tuning. Noncommercial end users are explicitly excluded.
Section 2, adding new 17 U.S.C. 514(a)(3) and (a)(5)
Why it matters:This is standard legislative practice for defining terms. However, the specific wording here is crucial for determining the bill's reach and who is subject to subpoenas.
Case for: Supporters would argue these definitions are carefully crafted to target the entities most responsible for the creation and evolution of generative AI models and their training data, while exempting casual users. This ensures accountability rests with those who have the capacity and control over the training process.
Case against: Critics might argue that the definition of 'substantially modify' could be broad enough to encompass various levels of model adaptation, potentially extending the subpoena burden to entities that perform minor adjustments or fine-tuning, rather than just core developers. The exclusion of 'noncommercial end users' is clear, but the line for 'developer' could be debated.
Estimated impact: This impacts potentially thousands of entities involved in AI development, from large tech companies to smaller startups and academic institutions, by defining who is subject to the new subpoena power.