Why It Matters
Federal courts have reached different conclusions about whether artificial intelligence (AI) companies can use copyrighted works to train their models without permission, leaving Congress to decide whether to establish new legal protections for copyright holders or let courts continue interpreting existing law. A Congressional Research Service (CRS) report examines the unresolved legal questions surrounding AI training and copyright ownership, including whether training qualifies as fair use and whether works created with AI assistance are eligible for copyright protection.
The Trump administration has favored allowing the courts to resolve the training question rather than having Congress establish new restrictions. The White House's National Policy Framework for Artificial Intelligence, released March 20, states that the administration believes training AI models on copyrighted material does not violate copyright law but acknowledges opposing legal arguments. It recommends that Congress avoid legislation that would interfere with the judiciary's resolution of the fair-use question.
The administration has also intervened in litigation. In September, the Department of Justice filed a brief supporting OpenAI's position in a copyright dispute, reinforcing the administration's approach to AI development and intellectual property protections.
The Big Picture
The legal uncertainty has fueled extensive litigation, with CRS citing a survey that identifies more than 150 copyright lawsuits against AI companies in the United States.
Two federal district court decisions issued in June 2025 illustrate the competing legal interpretations. In Bartz v. Anthropic and Kadrey v. Meta Platforms, judges in the Northern District of California concluded that the AI training uses before them qualified as fair use. However, the decisions differed in how they treated potential market harm. The courts examined whether AI-generated material could compete with copyrighted works and whether that competition should weigh against a fair-use defense.
In Bartz, the court also distinguished between using lawfully obtained books for training and Anthropic's acquisition and retention of pirated copies. Anthropic subsequently agreed to pay $1.5 billion to settle claims involving its use of pirated books, and the class settlement received final approval July 20.
A separate appellate decision reached a different result. On Sept. 29, the U.S. Court of Appeals for the Third Circuit ruled that ROSS Intelligence's copying of copyrighted Westlaw case summaries to train an AI-powered legal research system was not fair use. The court found that ROSS used the material to develop a competing commercial product, that the copying was minimally transformative and that it harmed the potential market for licensing Westlaw's copyrighted content.
The decision could influence future AI copyright litigation, although its application to generative AI may be limited. Unlike systems that generate original text or images, ROSS developed a legal research platform designed to identify relevant judicial opinions. The Third Circuit expressly distinguished its decision from cases involving generative AI models. Courts have also addressed whether AI-generated material can receive copyright protection.
In Thaler v. Perlmutter, the U.S. Court of Appeals for the District of Columbia Circuit held that copyright protection requires human authorship, rejecting an attempt to register a work attributed entirely to an AI system. The Supreme Court declined to review the decision March 2, leaving the appellate ruling in place.
The U.S. Copyright Office's 2023 guidance similarly maintains that AI-generated material is not independently eligible for copyright protection, although works containing AI-generated elements may qualify when they include sufficient original human authorship.
Congress is considering several proposals addressing different aspects of AI training, copyright liability and transparency. Five measures introduced during the 119th Congress have been referred to committee:
- S. 1396, the COPIED Media Act, would establish civil liability related to certain uses of copyrighted material in AI training. The legislation was referred to the Senate Commerce, Science, and Transportation Committee April 9, 2025.
- S. 2367, the AI Accountability and Personal Data Protection Act, would establish additional civil liability provisions involving AI systems and protected information. It was referred to the Senate Judiciary Committee July 21, 2025.
- S. 2455, the TRAIN Act, would allow copyright owners to seek subpoenas requiring AI developers to disclose whether their copyrighted works were used to train AI models. Its House companion, H.R. 7209, was referred to the House Judiciary Committee Jan. 22.
- S. 3813, the CLEAR Act, would require disclosures to the Copyright Office concerning copyrighted works used to train AI systems. It was referred to the Senate Judiciary Committee Feb. 10.
The proposals reflect competing approaches to AI copyright policy. Some would give copyright holders additional legal remedies or access to information about training datasets, while others would establish new disclosure obligations for AI developers.
The White House has taken a different approach, recommending that Congress allow the courts to determine whether AI training constitutes fair use. Its policy framework also suggests exploring collective licensing arrangements that could allow copyright holders to negotiate compensation from AI companies without violating antitrust laws.
The Bottom Line
Congress faces a choice between establishing new rules governing copyrighted material used in AI training and allowing federal courts to develop those rules through individual lawsuits. The Third Circuit's September ruling against ROSS Intelligence provides an appellate precedent rejecting a fair-use defense for certain AI training activities, but it does not resolve the broader question of whether generative AI companies can lawfully train their models on copyrighted works without permission.
Meanwhile, the separate question of copyright ownership remains governed by the human-authorship requirement, with courts and the Copyright Office distinguishing between independently generated AI material and works containing original human contributions. Another ruling on AI-assisted authorship could emerge from Allen v. Perlmutter, where the plaintiff and the Copyright Office have filed competing motions for summary judgment.
None of the five congressional measures identified by CRS has advanced beyond committee referral. Until Congress acts or additional appellate decisions establish broader precedents, the legality of AI training on copyrighted material will continue to depend on the circumstances of individual cases.
Access the Legis1 platform for comprehensive political news, data, and insights
Spot something wrong? Report an issue with this article