Written by: Cassie Turczyn
Artificial intelligence litigation is relatively new. Cases began to surface within the past five or six years, with complaints such as Thomson Reuters v. Ross Intelligence Inc. being filed in 2020. While this topic feels both new and like ‘old news,’ courts have been reluctant to adopt an all-encompassing rule pertaining to the rights associated with AI, especially in the realm of copyright protection. As of today, there is no presiding rule across circuits pertaining to AI usage of copyrighted materials for training purposes. Each circuit approaches this highly discussed topic a little differently, and most districts underneath have yet to grapple with cases in this area—at least for right now.
What are courts looking for?
Courts are considering all four § 107 factors under the Copyright Act of 1976, including the purpose or character of use, nature of the copyrighted work, amount and substantiality of what was used, and market effect, with the generative versus non-generative AI distinction being an almost dispositive variable alongside the potential market harm. The current analysis involves the “transformative” use or purpose, mainly governed by Andy Warhol Foundation v. Goldsmith, the framework has “cabined transformative use,” by holding that where a secondary use shares the same or highly similar purpose as the original and is commercial, “the first factor is likely to weigh against fair use, absent some other justification for copying.” Google v. Oracle articulated that fair use must adapt to “significant changes in technology” and that a highly transformative purpose “can justify copying.”
Third Circuit
Only one circuit—the Third Circuit—has ruled on fair use for AI training purposes, and that was decided in Ross on September 29, 2026. In this case, Thomson Reuters sued the AI startup Ross Intelligence for copyright infringement, claiming that it had unlawfully copied its Westlaw headnotes to build an AI-powered legal search engine to directly compete with its service. This occurred after Ross asked to license the tool from Thomson Reuters, but they declined. On factor one, the court held Ross’ use was “minimally transformative, at best” because both Thomson Reuters and Ross used the headnotes for the same ultimate purpose: powering a legal research platform. Unlike the intermediate copying cases in Oracle, where copying was necessary to access unprotected functional elements of computer code, ROSS copied the headnotes merely because it was easier than using those that were freely available. The Warhol framework was applied to disfavor fair use when the secondary use shares a highly similar commercial purpose to the original.
Through ROSS, the Third Circuit affirmed the district court’s holding “that the 2,243 Westlaw headnotes at issue were original enough for copyright protection and that ROSS’s copying of these headnotes did not constitute fair use” (Stanford Libraries Copyright and Fair Use). The circuit judge went further to hold that “ROSS’s copying and use of the headnotes to train an AI legal search platform was not fair use, given the highly commercial nature of ROSS’s use, its minimal transformative purpose, the substantial copying involved, and the harm to both the value and potential markets for Thomson Reuters’s work.”
Other Circuits
The Second Circuit and the Ninth Circuit usually are markers for the newest analysis in copyright law, due to them covering New York and California. That’s why it’s so surprising that the Third Circuit, spanning Delaware, New Jersey, Pennsylvania, and the U.S. Virgin Islands, was the first to take a definitive stance.
Both the Second and Ninth Circuits’ appellate decisions are silent on this fair use issue, but both have similar pending litigation that may soon reach the appellate courts. Of note, these include In re OpenAI, Inc. (2d Cir.) and Kadrey v. Meta (9th Cir.).
In In re OpenAI, Inc. in the Second Circuit, the Southern District of New York held that the complaint “squarely alleges” that OpenAI had access to plaintiffs’ works and that ChatGPT’s allegedly infringing outputs are based on plaintiffs’ works, satisfying the “actual copying” element of a prima facie infringement claim. The court then concluded that a reasonable jury could find that the detailed summary is substantially similar to the plaintiff’s original work to find improper appropriation through conveying the “overall tone and feel” of the original work, which is thus sufficient to defeat OpenAI’s motion to dismiss. This case is still in discovery and may go to trial, bringing us closer to the Second Circuit’s answer to the fair use defense in LLM training.
In Kadrey v. Meta in the Ninth Circuit, the Northern District of California held that Meta’s actions were protected by fair use and thus, did not constitute a copyright infringement claim. It reasoned by weighing the four factors for fair use: the purpose and character of training a large language model is highly transformative because a tool that generates summaries or derivative stories is fundamentally different from reading the book as written, the original work was highly creative, complete copying was necessary to achieve the transformative purpose, and there was no evidence that the outputs acted as substitutes for the original books or created market harm.
Significantly, Bartz v. Anthropic PBC held that training AI on copyrighted books constituted fair use, while storing entire pirated copies for retrieval is not protected. This case is in the same district court as Kadrey within the Ninth Circuit. In Bartz, Anthropic trained AI models on books from pirate sites and also maintained a library of the infringing works for uses beyond training. The court held that the training itself was fair use, but the library was not protected. In both cases, the court emphasized the transformative purpose of the new use, rather than how the company obtained the materials used for training purposes.
What is the Future of Generative AI Fair Use?
Based on the district court rulings, it’s hard to say whether there will be a circuit split at the appellate level. However, it is highly likely that litigation pertaining to AI companies’ use of copyrighted materials will rapidly increase as large language models are becoming commonplace in classrooms, law firms, and corporations. Regulations may likely come from Congress once these issues pan out, but the courts may hesitate to draw a universal approach due to the evolving nature of the technology and its effect on society.
This issue will likely be addressed by the Supreme Court on appeal and may be a close decision. Artificial intelligence could pose a threat to creative expression, independent thought, professional motivation, and natural resources, but it also could raise major legal discrepancies based within fair use doctrine and beyond. However, there are so many reasons to protect artificial intelligence, especially in a democratic country built on the marketplace of ideas that LLMs can tap into with the click of a button: progress of science, ease of mundane tasks, deeper understanding of complex topics, and responsible creation of AI-assisted works.
Sources:
17 U.S.C. § 107 (2018).
A.V. ex rel. Vanderhye v. iParadigms, LLC, 562 F.3d 630 (4th Cir. 2009).
Andy Warhol Found. for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508 (2023).
Authors Guild v. Google, Inc., 804 F.3d 202 (2d Cir. 2015).
Bartz v. Anthropic PBC, 787 F. Supp. 3d 1007 (N.D. Cal. 2025).
Fair Use and AI Training, Dave ADR, https://daveadr.com/blog/fairuseandaitraining (last visited Oct. 14, 2026).
Google LLC v. Oracle Am., Inc., 593 U.S. 1 (2021).
Huckabee v. Bloomberg L.P., No. 1:23-cv-09152, 2025 WL 3277560 (S.D.N.Y. Nov. 24, 2025).
In re OpenAI, Inc., Copyright Infringement Litig., 800 F. Supp. 3d 602 (S.D.N.Y. 2025).
Intercept Media, Inc. v. OpenAI, Inc., 767 F. Supp. 3d 18 (S.D.N.Y. 2025).
Kadrey v. Meta Platforms, Inc., 788 F. Supp. 3d 1026 (N.D. Cal. 2025).
Lehrman v. Lovo, Inc., 790 F. Supp. 3d 348 (S.D.N.Y. 2025).
Loeb & Loeb LLP, In re OpenAI, Inc. Copyright Infringement Litigation, Loeb & Loeb (Nov. 2025), https://www.loeb.com/en/insights/publications/2025/11/in-re-openai-inc-copyright-infringement-litigation (last visited Oct. 14, 2026).
N.Y. Times Co. v. Microsoft Corp., 777 F. Supp. 3d 283 (S.D.N.Y. 2025).
Perlmutter v. Blanche, No. 25-5285, 2025 WL 2627965 (D.C. Cir. Sept. 10, 2025).
Thomson Reuters Enter. Ctr. GmbH v. ROSS Intel. Inc., No. 25-2153, 2026 WL 2932557 (3d Cir. Sept. 29, 2026).
Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc., Stan. Univ. Libr.: Copyright & Fair Use, https://fairuse.stanford.edu/case/thomson-reuters-enterprise-centre-gmbh-v-ross-intelligence-inc/ (last visited Oct. 14, 2026).