Rapid advances in artificial intelligence (AI) are forcing courts to address questions that existing copyright laws are ill-equipped to answer. Large language models and generative AI tools depend on ingesting vast amounts of copyrighted material to learn and operate effectively. This raises a fundamental legal dilemma: does using copyrighted works to train AI models qualify as fair use, or is it infringement that requires permission and payment?

Recent court decisions have handed AI developers early wins, finding that training models on copyrighted content can be “transformative” and protected under fair use. But these rulings are narrow, fact-specific, and confined to the parties before the court. They remain vulnerable on appeal and do little to provide lasting clarity for an industry racing ahead of the law. History shows that when disruptive technologies collide with outdated legal frameworks, courts alone are rarely able to craft comprehensive solutions. The AI copyright battles currently unfolding are only the beginning and, ultimately, it will likely be up to Congress or the Supreme Court to resolve these unprecedented challenges.

Early Judicial Rulings on AI and Copyright

These tensions came into sharp focus late last month, when two federal judges in California handed AI companies victories in the ongoing debate over whether they can train their models on copyrighted materials without permission. In Bartz v. Anthropic PBC¹ and Kadrey v. Meta Platforms, Inc²., the courts addressed whether using copyrighted books to train large language models (LLMs) constitutes fair use under Section 107 of the Copyright Act.

Both cases turned on the familiar four factors: (1) whether the use is transformative; (2) the nature of the original works; (3) the amount and significance of the portion used; and (4) the impact on the market for those works. Each court primarily focused on the first and fourth factors, examining the transformative nature of the AI training and whether it harmed the market for the copyrighted materials. Both defendants won on summary judgment, but the judges took notably different approaches.

In Bartz, authors sued Anthropic for copying millions of books without permission to train its language model. Judge Alsup found that using lawfully purchased copies to extract statistical relationships for AI training is “exceedingly transformative” and constitutes fair use. He also found that digitizing print books for searchable reference is fair use. However, the court drew a firm line at piracy, allowing claims involving pirated copies to proceed to trial. Judge Alsup also criticized Anthropic for retaining a subset of pirated books in its central library even though they were not used for training, making clear that while transformative use can support fair use, outright piracy does not.

Similarly, in Kadrey, authors sued Meta for using books from unauthorized “shadow libraries” to train its LLaMA model without authorization. While Judge Chhabria agreed that using copyrighted works for AI training is “highly transformative,” he underscored that transformation alone is not dispositive. Instead, taking aim at Bartz, he emphasized market harm as the most important factor in fair use analysis. Although he ultimately ruled for Meta, Judge Chhabria stressed his decision was narrow, noting that the plaintiffs made the “wrong arguments” and failed to build a record in support of the right arguments, suggesting that future cases with stronger evidence of market harm may lead to different outcomes.

Meanwhile, there are numerous ongoing litigations in various district courts, posing the question of what constitutes “fair use” when purveyors of AI products glean information from published material to deploy as “learning” for these tools. While many such cases have settled over the past year or two, many more continue to wind through the system. Some of these disputes have significant economic issues at stake and one or more of these hotly contested disputes are widely expected to reach the U.S. Supreme Court.

And as if that was not enough to engender confusion, as late as July 23, 2025, President Trump has weighed in on the side of wanting to declare all training access as “fair use.” In a speech at the “Winning the AI Race” summit in Washington, Trump called for a “common sense” approach – “You can’t be expected to have a successful AI program when every single article, book or anything else that you’ve read or studied, you’re supposed to pay for. You just can’t do it, because it’s not doable.” Coming down, in this instance, in a way that sounds a bit like Judge Alsup’s ruling in Bartz, this is not likely to be the last word on the subject.³ Also on July 23, 2025 the White House issued a comprehensive AI policy document – “Winning the Race:  America’s AI Action Plan” – designed to define actions calculated to advance commercialization of AI technologies, moderate regulation and improve the U.S. position in the global sector.⁴ Earlier in the same week, United States Senators Hawley of Missouri and Blumenthal of Connecticut introduced legislation that would move in the opposite direction on training, and condition all such uses on obtaining permission from content owners.⁵ Clearly, there is a lot going on. 

Historical Lessons: Why Courts Alone Are Not Enough

As illustrated above, the broader AI-copyright landscape remains unsettled and is likely to continue evolving in a fragmented manner as courts issue varied rulings. Indeed, history shows that when disruptive technologies collide with existing copyright frameworks, lower courts alone rarely resolve the systemic challenges. Ultimately, where the judiciary can only provide partial solutions, Congress or the Supreme Court has often been compelled to step in to create durable solutions.

Consider the nearly decade-long Google Books litigation (Authors Guild, Inc. v. Google Inc.),⁶which ended with the Second Circuit ruling that Google’s scanning of millions of books and display of brief snippets was fair use because it enabled search and discovery without supplanting the originals works. Although the plaintiffs appealed to the Supreme Court, the Court declined to hear the case, leaving the Second Circuit’s ruling in place. While this decision supported technological innovation, it left unresolved broader policy questions about mass digitization, data use, and access to knowledge. These issues remain relevant as AI developers today aggregate vast datasets for training.

Similarly, when the Ninth Circuit shut down Napster, a peer-to-peer file-sharing network, for copyright infringement in the early 2000s, piracy did not disappear.⁷ Unauthorized audio file sharing continued to flourish despite those rulings, demonstrating that litigation alone could not stem widespread technological disruption without broader legislative or industry-based solutions. Congress attempted a partial fix through the Digital Millenium Copyright Act (DMCA), but piracy persisted.

By contrast, Sony Corp. v. Universal City Studios⁸demonstrates how Supreme Court intervention can provide decisive guidance. Confronted with fears that videocassette recorders (VCRs) would devastate the film and television industries, the Court ruled that “time-shifting,” which involved recording broadcasts for later personal viewing, was fair use. The Court recognized that copyright law must adapt to technological change and ruled that manufacturers of devices with substantial non-infringing uses were not liable simply because their products could also enable infringement. In doing so, Sony bridged traditional copyright principles with emerging technologies, resolving a systemic conflict and paving the way for the booming home video industry.

However, not all Supreme Court interventions have provided such clarity. For example, in ABC, Inc. v. Aereo, Inc.,⁹ the Supreme Court held that Aereo’s internet-based retransmission of over-the-air television broadcasts was not fair use and effectively treated Aereo as a cable provider requiring retransmission consent. While this decision ended Aereo’s business model, it left unresolved how copyright law applies to cloud-based broadcast technologies more broadly, creating continued uncertainty for future digital delivery innovators. Notably, the Court acknowledged that Congress was better positioned to craft a comprehensive solution for such emerging technologies.

Often, it has been Congress that has stepped in to address gaps the judiciary could not fill. For example, in the 1990s, the perfect digital copying made possible by digital audio tape (DAT) technology threatened the music industry’s revenue model. Courts could not resolve this conflict under existing doctrines, leading Congress to enact the Audio Home Recording Act to balance consumer recording rights with royalties for rights holders. Likewise, after Stratton Oakmont, Inc. v. Prodigy Services Co.¹⁰threatened to treat online platforms as publishers liable for user-generated content, Congress passed Section 230 of the Communications Decency Act, shielding platforms from most liability and enabling the internet to flourish as an open communications forum.

These examples illustrate a consistent pattern: while courts play a critical role in shaping doctrine and resolving disputes between parties, when new technologies create widespread legal uncertainty and threaten existing business models, only Congress or the Supreme Court can provide the comprehensive, durable solutions needed. As AI continues to challenge copyright law in fundamental ways, history suggests that piecemeal judicial decisions will not be enough. Eventual comprehensive legislative or judicial action is necessary.¹¹

1. Bartz v. Anthropic PBC, No. 24-cv-05417 (N.D. Cal. June 23, 2025).↩
2. Kadrey v. Meta Platforms, Inc., No. 23-cv-03417 (N.D. Cal. June 25, 2025). ↩
3. See stories reported in Daily Variety, July 23, 2025 (variety.com/2025/politics/news/trump-rejects-ai-training-compensation-1236468459) and Wall Street Journal, July 24, 2025, p.A1-“New Plan Aims to Buoy U.S. in Global AI Race”.↩
4. See “White House Unveils New AI Action Plan” July 24, 2025.↩
5. https://www.hawley.senate.gov/hawley-blumenthal-unveil-bipartisan-bill-empowering-working-americans-to-sue-big-tech-ai-companies-for-stealing-creative-works/ (last visited July 24, 2025). ↩ 
6. Authors Guild v. Google, Inc., 804 F.3d 202 (2d Cir. 2015). ↩
7. A&M Records, Inc. v. Napster, Inc., 239 F.3d 1004 (9th Cir. 2001).↩
8. Sony Corp. of Am. v. Universal City Studios, Inc., 464 U.S. 417 (1984). ↩
9. American Broadcasting Companies, Inc. v. Aereo, Inc., 573 U.S. 431 (2014). ↩
10. Stratton Oakmont, Inc. v. Prodigy Services Co., 1995 WL 323710 (N.Y. Sup. Ct. May 24, 1995).↩
11. Thanks to Dentons Summer Associate Olivia Bartell, Loyola University Chicago, School of Law, Class of 2026, for her invaluable assistance.↩