Congresswoman Madeleine Dean and Congressman Nathaniel Moran this week introduced bipartisan legislation establishing an administrative subpoena process enabling copyright owners to determine whether their creative works were used to train generative artificial intelligence models. The Transparency and Responsibility for Artificial Intelligence Networks Act grants musicians, artists, writers, and other creators legal mechanisms to access training records from AI developers who may have used copyrighted material without authorization.

According to the press release, the legislation addresses a fundamental gap in current law. No process currently exists for creators to determine if generative AI models use their work without consent or compensation to train systems. The TRAIN Act models its approach after legal procedures used for internet piracy cases.

The bill establishes specific requirements for both copyright owners seeking information and AI developers subject to subpoenas. Copyright owners must demonstrate subjective good faith belief that developers used their copyrighted works for training purposes. They file proposed subpoenas with district court clerks alongside sworn declarations explaining their purpose involves protecting copyright holder rights.

According to the bill text, developers subject to subpoenas must expeditiously disclose requested information. Non-compliance creates a rebuttable presumption that the developer made copies of the copyrighted work. This provision establishes consequences for AI companies that refuse to provide transparency about their training datasets.

The legislation defines developers broadly to include entities that design, code, produce, own, or substantially modify generative AI models. According to Section 514(a)(3) of the proposed statute, the definition encompasses third-party training dataset curators who engage in or supervise dataset curation or use training datasets for model training. Noncommercial end users remain explicitly excluded from coverage.

Training material specifications include individual works or components used for generative AI model training purposes. According to the statutory language, this encompasses combinations of text, images, audio, or other categories of expressive materials, along with annotations describing the material. Subpoenas are limited to works owned or controlled by the requesting copyright holder.

The introduction marks the first time the TRAIN Act reaches the House of Representatives. Senators Peter Welch, Marsha Blackburn, Adam Schiff, and Josh Hawley reintroduced the legislation in the Senate in July 2025, establishing initial framework for congressional consideration. The House version mirrors Senate provisions while expanding legislative support across both chambers.

Representative Dean emphasized the necessity of updating legal frameworks to address AI's rapid development. "As AI rapidly evolves and becomes more present in our lives, our laws must catch up - that includes preserving the dignity of artists and the authenticity of their work," Dean stated in the press release. She highlighted that no path currently exists for creators to know if their work has been used without permission and compensation to train AI models.

Representative Moran characterized artificial intelligence as a powerful engine of creativity and innovation for the American people. "It expands opportunity, drives progress, and strengthens American leadership," Moran stated. He emphasized that transparency is essential to ensuring innovation is built on integrity and respect for original work.

The legislation arrives amid intensifying legal challenges to AI training practices. Anthropic PBC agreed to pay at least $1.5 billion in September 2025 to settle copyright infringement allegations after authors claimed the company illegally used pirated copies of books to train large language models. According to Justin Nelson, representing the authors, the settlement constitutes the largest publicly reported copyright recovery in history.

Court decisions have produced mixed outcomes on fair use applications to AI training scenarios. Meta won summary judgment in June 2025 when Judge Vince Chhabria ruled that using copyrighted books to train Llama large language models constituted fair use because the usage was highly transformative. However, that ruling applied only to specific plaintiffs, leaving other authors whose works were used in Llama training free to pursue their own copyright claims.

The US Copyright Office released major guidance in May 2025 addressing when AI developers need permission to use copyrighted works. The 107-page report provided exhaustively detailed analysis of how generative AI development implicates copyright law and when the fair use doctrine may apply. Rather than making sweeping determinations, the Office established a nuanced framework for case-by-case evaluation.

The procedural framework outlined in Section 2 of the TRAIN Act requires copyright owners to file with district court clerks both a proposed subpoena and a sworn declaration. The declaration must state that the legal or beneficial owner has subjective good faith belief that the developer used some or all of one or more copyrighted works to train the generative artificial intelligence model.

According to subsection (c)(2)(B), the purpose for which the subpoena is sought must be to obtain copies of training material, or records sufficient to identify with certainty the training material used to train the generative artificial intelligence model. This enables copyright owners to determine whether developers have used copyrighted works in connection with the generative artificial intelligence model.

If a proposed subpoena is in proper form and the accompanying declaration is properly executed, the clerk must expeditiously issue and sign the proposed subpoena and return it to the requester for delivery to the developer. This administrative process eliminates the need for full litigation before copyright owners can access basic information about potential infringement.

The legislation includes safeguards against abuse through sanctions provisions. According to subsection (j), if a copyright owner requests a subpoena in bad faith, the court that issued the subpoena may impose sanctions on the legal or beneficial owner or authorized person upon motion of the subpoena recipient. Rule 11(c) of the Federal Rules of Civil Procedure applies to sanctions under this provision.

Duty of confidentiality provisions require that legal or beneficial owners who receive copies or records from developers may not disclose them to any other person without proper authorization or consent. This protection addresses AI companies' concerns about protecting proprietary training methodologies while enabling copyright enforcement.

The bill establishes technical definitions for artificial intelligence and related concepts. According to Section 514(a)(1), artificial intelligence carries the meaning given in Section 5002 of the National Artificial Intelligence Initiative Act of 2020. Generative artificial intelligence models are defined as systems that emulate the structure and characteristics of input data to generate derived synthetic content including images, videos, audio, text, and other digital content.

Substantially modify receives specific definition as taking one or more actions leading to new versions, new releases, or other updates to generative artificial intelligence models that materially change functionality or performance. According to the statutory language, this includes retraining or fine tuning the generative artificial intelligence model.

The legislation received endorsements from multiple creative industry organizations. The Recording Industry Association of America applauded Representatives Dean and Moran for their leadership on the bipartisan, bicameral TRAIN Act. "This commonsense legislation ensures that artists and rights holders have meaningful access to the courts when their work is copied for AI training without authorization or consent," stated Mitch Glazier, RIAA Chairman & CEO.

SAG-AFTRA commended the representatives for leading with clarity and courage at a pivotal moment for creativity. "Transparency is not a barrier to progress; it's a bridge that ensures A.I. advances alongside the artists and workers whose creativity powers our culture," stated Duncan Crabtree-Ireland, SAG-AFTRA National Executive Director & Chief Negotiator.

The Recording Academy characterized the TRAIN Act as empowering creators with an important tool to ensure transparency and prevent misuse of copyrighted works. "The Recording Academy applauds Representatives Dean and Moran for their leadership and commitment to protecting human creators and creativity," stated Todd Dupler, Recording Academy Chief Advocacy & Public Policy Officer.

The Human Artistry Campaign described the legislation as an important step toward effective AI regulatory regime that respects creators' rights. "Transparency is a key tenet of the Human Artistry Campaign's principles for responsible and ethical AI, and we're grateful to Representatives Dean and Moran for leading on this issue," stated Dr. Moiya McTier.

The American Association of Independent Music emphasized that generative AI is already disrupting the independent music sector. "Music creators have no actionable way of knowing which models are built off their intellectual property. The TRAIN Act would fill that void in the regulatory landscape and serve as a pathway for A2IM members to enforce their rights," stated Ian Harrison, CEO of A2IM.

The American Federation of Musicians stressed the necessity of transparency for working musicians. "It is essential that all artists know when our work is used for machine learning. We must have the ability to protect our livelihoods. The TRAIN act will end the guessing game and create transparency where it is desperately needed," stated Tino Gagliardi, President.

Copyright Clearance Center characterized transparency as promoting development of reliable, ethical, and trustworthy AI systems. "Unfortunately, it is exceedingly difficult for copyright owners and artists to know if and how AI systems have used their work for training purposes. The TRAIN Act would provide a path to information sharing without the need to resort to infringement lawsuits," stated Tracey Armstrong, CEO.

Global Music Rights emphasized that transparency, equity, and protection of creators' rights form the foundation of everything they do. "In the rapidly evolving landscape of the music industry, human creators should not only be properly compensated - they must have autonomy. And this legislation is a critical step toward safeguarding those essential rights," stated Emio Zizza, General Counsel.

The International Alliance of Theatrical Stage Employees highlighted that professional crews powering film and television production depend on AI tools development that respects and pays for human creativity. "Transparency efforts like the TRAIN Act mark a first, but absolutely critical, step towards ensuring that the production workforce continues to benefit from how their labor is used in the marketplace," stated Matthew D. Loeb, International President.

Nashville Songwriters Association International emphasized that human authors and their copyrights must be valued and protected. "The TRAIN Act will create much-needed transparency around Generative Artificial Intelligence that, unchecked, threatens the livelihoods of human creators. We are appreciative to Reps. Dean and Moran for their leadership on this issue," the organization stated.

The International Association of Scientific, Technical and Medical Publishers expressed support for mechanisms promoting transparency. "AI technologies present scientists and publishers with new tools to enhance discoveries, maintain the integrity of the scientific record, and accelerate breakthroughs," stated Dr. Caroline Sutton, CEO. The organization emphasized that scientific knowledge requires a traceable history shaped by human discovery, transparent analysis, replicability, and rigorous editorial review.

The legislation addresses concerns that extend beyond traditional creative industries. Research published in January 2026demonstrated that Stanford University researchers successfully extracted large portions of copyrighted books from four production large language models. Claude 3.7 Sonnet reproduced 95.8% of Harry Potter and the Sorcerer's Stone nearly verbatim in a single experimental run, highlighting memorization risks despite safety measures.

The marketing technology sector faces increasing scrutiny around AI training data practices. LinkedIn expanded AI training to include user data starting November 3, 2025, introducing new privacy controls across multiple regions. The platform announced it would collect profile data, jobs-related information, member content including posts and articles, and generative AI usage data for model training purposes.

Copyright litigation intensifies as multiple AI companies face challenges from content creators. Ziff Davis filed a major lawsuit against OpenAI in April 2025, accusing the company of unauthorized use of content from 45 properties including CNET, IGN, and Mashable. Reddit sued Anthropic in June 2025 over alleged unauthorized AI training on platform data.

Regulatory frameworks continue developing across international jurisdictions. The Dutch Data Protection Authority launched comprehensive consultation in May 2025 outlining how data protection laws should apply to generative artificial intelligence development and deployment. According to the authority, it is plausible that irregularities occurred during foundation models development, with the overarching estimate suggesting the vast majority of generative AI models currently fall short in terms of legitimacy.

The debate over copyright and AI training reflects fundamental questions about intellectual property rights in the AI era. Andreessen Horowitz submitted formal comments to the US Copyright Office in October 2023 arguing that using copyrighted content to train AI models constitutes fair use under existing law. The firm stated that AI model training extracts statistical patterns and facts rather than storing copyrighted content.

Critics have challenged these positions. Kristen Ruby disputed Andreessen Horowitz's interpretation in January 2025, arguing that AI models do store copyrighted content during training. Ruby contested the characterization of training data use as statistical facts extraction, maintaining that copyright infringement occurs regardless of how the data is processed.

The effective date provisions establish that Section 514 takes effect on the date of enactment. This creates immediate applicability once the President signs the legislation into law. The technical and conforming amendment adds the new section to the table of sections for Chapter 5 of Title 17, United States Code.

Congresswoman Dean has been a leader in Congress for AI regulation. Last year, President Trump signed her TAKE IT DOWN Act into law, becoming the first federal legislation to address AI by requiring social media platforms to remove real and AI-generated non-consensual intimate imagery.

Representative Moran represents Texas's First Congressional District. The bipartisan nature of the legislation reflects growing consensus across political divides that creators require tools to protect their intellectual property rights in the age of generative AI.

The procedural approach mirrors mechanisms used for Digital Millennium Copyright Act subpoenas targeting internet piracy. Copyright owners could use similar clerk-issued subpoenas to identify internet service provider subscribers allegedly sharing copyrighted files. The TRAIN Act adapts this framework to address AI training data opacity.

The legislation does not establish whether AI training on copyrighted works constitutes infringement. It merely provides discovery mechanisms enabling copyright owners to determine if their works were used. Subsequent enforcement actions would proceed through normal copyright infringement litigation if owners identify unauthorized uses.

Industry observers note that transparency requirements could influence AI development practices. Companies might implement more rigorous documentation of training data sources to respond efficiently to subpoenas. Some developers may shift toward licensing arrangements or use of public domain materials to avoid disclosure obligations.

The advertising technology sector relies increasingly on generative AI for content creation, audience targeting, and campaign optimization. Marketing professionals using AI tools for creative generation, customer service, or data analysis must consider whether underlying models trained on copyrighted materials without authorization. The TRAIN Act would enable content creators whose works appear in marketing materials to investigate whether AI systems used their copyrighted content during training.

The timing of the House introduction coincides with mounting pressure on AI companies to address copyright concerns. Multiple lawsuits challenging training practices across federal courts have produced inconsistent rulings on fair use applications. The June 2025 Meta ruling found fair use in specific circumstances, while other courts have distinguished between legitimate training uses and pirated content acquisition.

Few AI companies currently share how their models are trained. Nothing in current law requires them to do so. The TRAIN Act would mandate disclosure when copyright owners meet the subjective good faith belief standard and follow proper procedural requirements.

The legislation establishes that unless otherwise provided by Section 514 or by applicable court rules, the procedure for issuance and delivery of subpoenas, and remedies for noncompliance, shall be governed to the greatest extent practicable by the Federal Rules of Civil Procedure governing subpoena duces tecum issuance, service, and enforcement.

Senators introducing the companion legislation emphasized the importance of establishing higher standards for transparency as AI becomes more embedded in daily lives. The bipartisan, bicameral nature of the TRAIN Act demonstrates unusual political alignment on intellectual property protection in the AI era.

The marketing community faces growing complexity around AI tool deployment as regulatory frameworks develop across multiple jurisdictions. European data protection authorities have established increasingly stringent requirements for AI systems processing personal data. The French data protection authority CNIL finalized recommendations in July 2025 requiring organizations to implement procedures for identifying individuals within training datasets and models.

Global copyright laws show signs of convergence regarding AI training uses. Research published in the Emory Law Journal in October 2024 revealed surprising alignment across jurisdictions including the United States, European Union, Japan, and China. Countries worldwide are increasingly embracing copyright exceptions to facilitate AI development while balancing creator protections.

The TRAIN Act represents a measured approach to addressing creator concerns without prohibiting AI development. By establishing transparency mechanisms rather than blanket restrictions, the legislation enables copyright enforcement while preserving space for legitimate transformative uses that courts may determine constitute fair use.

The administrative subpoena process reduces barriers to information access compared to full copyright infringement litigation. Copyright owners could obtain training data disclosure without needing to demonstrate actual infringement or survive motions to dismiss. This shifts burden to AI developers to either provide requested information or face the rebuttable presumption of copying.

The legislation received zero opposition statements in the press release, reflecting broad creative industry support for transparency measures. Organizations spanning music, film, television, publishing, and scientific fields endorsed the approach as appropriate balance between creator rights and technological innovation.

The House Judiciary Committee received referral of H.R. 7209 on January 22, 2026. Committee consideration will determine whether the legislation advances to floor votes. The Senate version introduced in July 2025 remained in committee as of the House introduction date.

Congressional action on AI regulation has accelerated as the technology's capabilities expand and commercial deployments proliferate. The TRAIN Act represents one element of broader legislative attention to artificial intelligence governance spanning safety, privacy, competition, and intellectual property concerns.

The advertising technology industry must monitor legislative developments as AI integration deepens across marketing platforms. Transparency requirements could affect how companies develop and deploy AI-powered tools for creative generation, audience analysis, and campaign optimization. Marketing professionals should evaluate whether AI systems they use for content creation, customer service, or data analysis trained on copyrighted materials without authorization.

The legislation establishes no prohibition on AI training using copyrighted works. It merely requires developers to disclose training data when copyright owners meet procedural requirements. Courts would determine in subsequent litigation whether specific training uses constitute fair use or require licensing.

Timeline

Summary

Who: Congresswoman Madeleine Dean (D-PA) and Congressman Nathaniel Moran (R-TX) introduced bipartisan legislation affecting copyright owners including musicians, artists, writers, and creators, alongside AI developers who design, code, produce, own, or substantially modify generative artificial intelligence models. Senate cosponsors include Peter Welch (D-VT), Marsha Blackburn (R-TN), Adam Schiff (D-CA), and Josh Hawley (R-MO). Endorsing organizations include RIAA, SAG-AFTRA, Recording Academy, Human Artistry Campaign, American Association of Independent Music, American Federation of Musicians, Copyright Clearance Center, Global Music Rights, IATSE, Nashville Songwriters Association International, and International Association of Scientific, Technical and Medical Publishers.

What: The Transparency and Responsibility for Artificial Intelligence Networks Act (H.R. 7209) establishes an administrative subpoena process allowing copyright owners to request district court clerks issue subpoenas to AI developers. The legislation requires developers to disclose training materials or records sufficient to identify copyrighted works used to train generative artificial intelligence models. Copyright owners must demonstrate subjective good faith belief that developers used their protected works for training purposes. Non-compliance creates rebuttable presumption that developers made copies of copyrighted works. The bill defines developers as entities designing, coding, producing, owning, or substantially modifying generative AI models, excluding noncommercial end users. Training material specifications include text, images, audio, and other expressive materials along with descriptive annotations.

When: The legislation was introduced January 22, 2026, in the House of Representatives. This marks the first time the TRAIN Act reaches the House after Senate introduction in July 2025. The bill takes effect on the date of enactment once signed into law.

Where: The legislation applies throughout the United States through federal district courts. Copyright owners file proposed subpoenas with district court clerks who issue subpoenas to AI developers subject to disclosure requirements. The House Judiciary Committee received referral of H.R. 7209 for consideration.

Why: The legislation addresses a fundamental gap where no process currently exists for creators to determine if generative AI models use their work without consent or compensation to train systems. Few AI companies currently share how their models are trained and nothing in current law requires them to do so. The TRAIN Act models its approach after legal procedures used for internet piracy cases to establish transparency mechanisms enabling copyright enforcement. The bill arrives amid intensifying legal challenges including Anthropic's $1.5 billion settlement and multiple lawsuits from publishers and content creators. Court decisions have produced mixed outcomes on fair use applications to AI training scenarios. Creative industry organizations spanning music, film, television, publishing, and scientific fields endorsed the approach as appropriate balance between creator rights and technological innovation.

Share this article
The link has been copied!