Authored By: Esan Oluwafeyikemi Kenechukwu
Babcock University
ABSTRACT
With the coming of generative Artificial Intelligence technologies, there has been a rise in alleged copyright infringement from various individuals, companies and organizations whose works have been used to train AI models, without their permission. At the core of these disputes lies the issue of fair use–whether the compilation of various copyrighted works into datasets to be fed into AI, can pass as copyright infringement under the law. U.S. and international courts are beginning to apply the four-factor fair-use test to AI training, in order to properly ascertain whether a particular claim is indeed copyright infringement. This article will examine prominent case law in this evolving area of law such as Andersen v. Stability AI, Thomson Reuters v. ROSS, and Bartz v. Anthropic, and take cognizance of the dichotomy of opinions regarding the transformative nature of AI.
INTRODUCTION
The use of generative Artificial Intelligence has skyrocketed rapidly in the last few years, with the likes of Stable Diffusion, Claude, and LLaMA transforming creative industries and content generation in ways people only dreamed of. But along with this unprecedented growth, there has been a rise in legal actions and claims specifically for copyright infringement. Since 2023, thirty copyright lawsuits have been initiated all over the world1 which have posed complex questions about the length of copyright protection in the age of AI.
At the crux of these legal battles is the doctrine of fair use. This is a loophole in the U.S copyright law that allows for the limited use of protected works without permission under certain conditions.2 Courts are now being met with a new challenge of applying the fair use four factor test to emerging AI systems that have been trained using large datasets, often consisting of copyright protected works, to produce synthetic content. This emerging jurisprudence on fair use with regards to AI technologies, will mark a doctrinal shift that could forever alter the application of copyright law not only in the U.S but internationally.
BACKGROUND
Generative AI is a type of artificial intelligence that creates new content such as songs, texts, audio, images, and videos based on information derived from training datasets.3 These AI technologies are trained using terabytes worth of datasets in order to accumulate large depths of information. Several generative AI technologies have emerged in the last few years with the likes of ChatGpt, Bard, Claude, DALL-E, Midjourney, Stable Diffusion, LLaMA, e.t.c. becoming an indispensable tool in our workplaces, schools as well as our entertainment. These models are built to identify patterns in training datasets and produce new results based on those patterns.
Fair use is a fundamental limitation on copyright owners’ exclusive rights, codified in American statute in Section 107 of the Copyright Act of 1976.4It is a legal doctrine, in copyright law, that allows a party to make use of a copyrighted work without permission from the copyright owner or the author. Normally, a person would need permission from an author or owner of a copyrighted work before they can make use of it. However, the fair use doctrine offers a bypass to this original order. The access given by fair use is restricted to certain portions of the copyrighted work in question in order not to put the author of a work at a disadvantage. Fair use serves as a line of defense against copyright infringement claims and is often applied for purposes like criticism, commentary, news reporting, teaching, scholarship, and research.5 But in order to ascertain what falls under the umbrella of fair use and what doesn’t, courts apply a four factor test that brings into consideration; (1) the purpose and character of the use, including whether such use is commercial or nonprofit educational; (2) the nature of the copyrighted work; (3) the amount and substantiality of the portion used in relation to the work as a whole; and (4) the effect of the use upon the potential market for or value of the copyrighted work.6
The fair use doctrine originated in the 18th-19th century common law to prevent copyright from being too overbearing and debilitating the very creativity which copyright law is designed to protect and promote. But the digital age brought an unprecedented pivot to the traditional understanding and application of the fair use doctrine. In the case of Authors Guild v. HathiTrust 7, the court deemed digitization of copyrighted works as under the fair use doctrine when the new use is adequately transformative. Now with generative AI on this rise, this doctrine will be further bent to its breaking point. The sheer scale and manner in which generative AI models take in large amounts of data-often compromising of several copyrighted works-to churn works that may or may not replicate original works, leaves many courts, legal practitioners and AI stakeholders to question whether the protection of the fair use can only stretch so far.
THE DICHOTOMY ON TRANSFORMATIVE USE: BARTZ V. ANTHROPIC PBC
In the case of Bartz v. Anthropic PBC 8, the Court in the northern district of California passed judgment in favor of Claude, a generative AI made by Anthropic. The court held that Anthropic’s use of copyrighted works to train its language model, Claude, was transformative enough to constitute fair use.9 The presiding judge made use of an analogy on human learning to put his decision into the right perspective. He likened AI training to that of a human being learning; just like how humans can read a copyrighted book and create another work based on the information obtained from that book, so also can AI rely on copyrighted works to gather information to produce new and creative results. The court pointed out the difference between the original works and the output from the AI model, noting that they were distinct from one another to constitute fair use.10 But this ruling sparked a debate within the legal sphere. The concept of transformative use was first opined in the case of Campbell v. Acuff-Rose Music,11 ever since it has become a yardstick in the analysis of fair use in the digital age. Transformative uses weigh heavily in favor of fair use as it can be argued that the original was used to create something completely different and unique. However, the US Supreme Court’s decision in Andy Warhol Foundation v. Goldsmith12 has further broadened this doctrine of transformative uses, stating that transformation must not be measured only by the sheer or technical differences between the original work and a new work, but it should also be taken into consideration whether the new work serves to attain a different market goal than the original.
Armed with these judicial precedents, many learned individuals believe that the ruling in Bartz v Anthropic was detrimentally superficial and failed to properly apply the doctrine of transformative use. Critics argue that the judge had failed to apply the principles established in Campbell v. Acuff-Rose Music and Andy Warhol Foundation v. Goldsmith. The decision in Bartz v Anthropic has been characterized by the Copyright Alliance as having “fatal flaws”in its transformative use analysis.13
COMMERCIAL SUBSTITUTION AS THE DECISIVE FACTOR: THOMSON REUTERS V. ROSS INTELLIGENCE
In a glaring disparity to the Bartz case, Judge Stephanos Bibas in Thomson Reuters v. ROSS Intelligence14 countered the fair use defense, emphasizing that ROSS’s AI tool directly competed with Thomson Reuters’ Westlaw service. The court found that ROSS’s use of Westlaw headnotes to train its legal AI research tool failed the transformation test as established by prior precedents because it “meant to compete with Westlaw by developing a market substitute”. This case followed the precedent set by Andy Warhol Foundation v. Goldsmith that a new work must have a different market function than the original.
Judge Bibas deliberately distinguished the case from favorable precedents like Google Books, noting that unlike Google’s search snippets, ROSS’s AI tool created “a rival product” that purposely operated in the same market as the original work. The court highlighted that the commercial nature of the competition was decisive: “making commercial use of vast troves of copyrighted works to produce expressive content that competes with them in existing markets, especially where this is accomplished through illegal access, goes beyond established fair use boundaries”.15
CONCLUSION
With the unprecedented arrival and growth of AI, many laws and legal doctrines are faced with the challenge of being refined to accommodate these technological developments or face redundancy. The law on fair use and AI remains disconnected but they seem to meet on core principles such as transformative use and market harm. Generative AI has permanently altered the trajectory of copyright law. Courts have now taken it upon themselves to refine the fair-use doctrine case by case, with market harm and transformative use as the focal point. This will serve as a wake up call to all AI pacesetters and stakeholders to be more rigorous in data hygiene towards training datasets, and be more open to negotiated licensing with authors of original works.
Reference(S):
1 Kupferschmid K, “Insights from Court Orders in AI Copyright Infringement Cases” (Copyright Alliance, December 12, 2024) https://copyrightalliance.org/ai-copyright-infringement-cases-insights/ accessed July 27, 2025
2 Noble T, “Two Courts Rule on Generative AI and Fair Use — One Gets It Right” (Electronic Frontier Foundation, June 26, 2025) https://www.eff.org/deeplinks/2025/06/two-courts-rule-generative-ai-and-fair-use-one-gets-it-right accessed July 27, 2025
3 Stryker C and Scapicchio M, “What Is Generative AI?” (Ibm.com, June 4, 2025) https://www.ibm.com/think/topics/generative-ai accessed July 27, 2025
4 Copyright Act 1976, s 107.
5 U.S. Copyright Office, “Fair Use (FAQ)” (Copyright.gov) https://www.copyright.gov/help/faq/faq-fairuse.html accessed July 27, 2025
6 “17 U.S. Code § 107 – Limitations on Exclusive Rights: Fair Use” (LII / Legal Information Institute) https://www.law.cornell.edu/uscode/text/17/107 accessed July 27, 2025
7 Authors Guild, Inc. v. HathiTrust, No. 12-4547 (2d Cir. 2014)
8 Bartz v. Anthropic PBC, 3:24-cv-05417, (N.D. Cal.)
9 “Bartz v. Anthropic: First Court Decision on Fair Use Defense in LLM Training – Wiggin and Dana LLP — Attorneys at Law” (Wiggin and Dana LLP — Attorneys At Law, June 30, 2025) https://www.wiggin.com/publication/bartz-v-anthropic-first-court-decision-on-fair-use-defense-in-llm-training/ accessed July 27, 2025
10 “AI Training and Fair Use: US Federal Court Delivers a Landmark Ruling in Bartz v. Anthropic” (Herzoglaw | Israeli Law Firm, July 17, 2025) https://herzoglaw.co.il/en/news-and-insights/ai-training-and-fair-use-us-federal-court-delivers-a-landmark-ruling -in-bartz-v-anthropic/ accessed July 28, 2025
11 Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569 (1994)
12 Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508 (2023).
13 Lang C, “Analysis in Bartz v. Anthropic AI Case Marred by Fatal Flaws” (Copyright Alliance, July 24, 2025) https://copyrightalliance.org/bartz-anthropic-ai-case-flaws/ accessed July 28, 2025
14 Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc., No. 1:20-cv-613-SB (D. Del.). 4
15 “U.S. Copyright Office Issues Final Report on AI” (www.ascap.com) https://www.ascap.com/news-events/articles/2025/05/copyright-office-ai-report-3 accessed July 28, 2025 5
