Rewarding Creativity or Condoning Copyright?

by Leah Luckett, Associate Member, University of Cincinnati Law Review Vol. 94

I. Introduction

Artificial intelligence models are quickly becoming a unique part of society. Every day, more members of society are turning to these models for answers and relying on the results. Although many do not think about where the information comes from, these models do not create answers out of thin air.¹ In actuality, AI models are trained and tested until they are proven accurate enough for use.²

The training involves feeding curated data into selected algorithms, allowing the system to refine itself and produce accurate responses to queries.³ In order to do this, large amounts of quality data must be gathered to train the model.⁴ Compiling such massive amounts of data requires AI developers to acquire information from a variety of sources, including the internet, academic journals, and commercial databases.⁵ Many pieces of authorship are included in this data, and are often acquired without the owner’s permission.⁶ What becomes difficult is determining whether AI developers must ask permission from every single source they use to help train the AI model. If so, this would create an enormous burden on the developers, as the added demands would increase not only the time involved but also the cost and efficiency of an already strenuous process.⁷ It could also lead to enormous amounts of contracts and paperwork, potentially discouraging companies from developing AI models.⁸

On the other hand, authors of news articles, books, plays, and other works are seeing their work product used without any recognition.⁹ AI models are fed these works and then spit out the information without crediting the original author.¹⁰ This issue has led to the emergence of many new lawsuits. In April 2025, a U.S. District Court consolidated 12 cases brought against OpenAI and Microsoft in California into a single case in New York.¹¹ Many opposed the consolidation because they were hoping to have their arguments heard on an individual basis, but it is clear there is one question the court is aiming to answer: are companies that use copyrighted works, without consent or compensation, to train their large language models that underlie generative artificial intelligence products, allowed to do so?¹²

This Article explores the strong arguments at play on both sides of the debate over whether the use of an author’s work to train AI models constitutes copyright infringement or fair use. Part II provides a background of the specific copyright issues and lawsuits that AI developers are facing. Then, Part III details the arguments of both sides and interprets past cases that may help predict how courts will rule. Finally, Part IV offers a brief conclusion on the issues as they stand today and ways the problem could be addressed outside of court.

II. Background

A. Introduction to Copyright Law and The Fair Use Doctrine

Copyright is a form of intellectual property that protects an original work of authorship once its author fixes the work in a tangible form of expression.¹³ To qualify for copyright protection, the work must be independently created by a human author and demonstrate a degree of creativity.¹⁴ Copyright law in the U.S. was instituted in 1790 by Congress and has been an integral part of protecting authorship ever since.¹⁵ Especially for the authors of books and articles, copyright provides protection for the strings of words and thoughts that are carefully crafted for their readers.¹⁶ Interestingly, copyright law was never intended to specifically protect authors, but rather to promote the progress of science and the useful arts.¹⁷ Copyright law has attempted to balance public interest in using works by allowing exceptions while also protecting the promotion of new works by authors to progress science and the arts.¹⁸ According to Congress, copyright law serves three purposes: the promotion of learning, the preservation of the public domain, and the protection of the author.¹⁹

This balance is reflected by the exceptions to copyright, which permit individuals to use another’s work without violating copyright law.²⁰ Because almost anything could be considered copyrighted, even when it is a new idea taken from previous renditions of works, the fair use doctrine evolved.²¹ Fair use allows material to be used without permission.²² Codified in federal law in 1976, the doctrine has become a central part of modern copyright law and allows academics, critics, journalists, teachers, film makers, and technology companies to rely on it in order to have a certain amount of freedom when dealing with other people’s copyrights.²³ It also allows courts to avoid rigid application of the copyright statute when it would stifle the very creativity the law is designed to foster.²⁴ However, the fair use doctrine is still narrowly construed and is not meant to override copyright laws or rules in the name of “fair use.”²⁵

Implementation of the fair use doctrine has led to ongoing debates as to what exactly fair use means, and questions about who can use copyrighted material and to what extent.²⁶ In 2005, Google undertook a large-scale, unauthorized digitization of library books to create an unashamedly commercial book search engine, Google Books.²⁷ Google was sued for violations of copyright laws, but argued that its conduct was protected under the fair use doctrine.²⁸ This case presented a question that is still lingering today: to what extent does fair use protect companies and users from unauthorized use of copyrighted work? In Google v. Author Guilds, the court dismissed the lawsuit and affirmed that Google Books met all legal requirements for fair use.²⁹ On appeal, the appellate court affirmed the district court’s ruling, holding that Google’s project provides a public service without violating intellectual property law.³⁰ This judgment was a major blow to those who had hoped for a narrower interpretation of fair use— one that did not protect massive companies from using authors’ works to their own benefit, without providing authors with the pay they demand.

B. The Fair Use Doctrine Today and its Application to AI Models

Today, the issue persists but in a different area: artificial intelligence models. Although some of the most mainstream AI models did not release until 2022, they have quickly picked up steam.³¹ Many Americans now use ChatGPT in their everyday lives, relying on AI for work, relationships, creativity, and a wide range of other purposes.

In 2022, these AI chatbots were less developed than they are now, just three
years later.³² Companies have learned where there are issues and have quickly resolved them, pushing AI to new limits.³³ This resolution has come with the incorporation of better information and new knowledge into the training that enables these AI models to become so brilliant.³⁴ AI models are trained continuously to become more efficient and more accurate in their answers.³⁵ Training an AI chatbot is done similarly by each company.³⁶ This training is done by a large language model (“LLM”) behind every AI tool.³⁷ The sheer volume of data makes it impossible for humans to label or explain it all. By exposure, the model learns grammar, facts, reasoning abilities, and even the biases present in the data.³⁸ The next part of the process is supervised learning.³⁹ This requires giving the LLM specific lessons and examples with carefully crafted examples of inputs and desired outputs.⁴⁰ Lastly, there is advanced fine-tuning, which relies on human judgment to evaluate and refine the models’ outputs to be as accurate and as helpful as possible.⁴¹

While this process seems simple, it is much more complicated in real life.⁴² Especially as authors argue for one additional concern that has not been accounted for—the overwhelming amount of data being fed to these models frequently involves copyrighted works that have not been authorized for use.⁴³

As of November 2025, there are more than 30 active lawsuits between large AI companies and creators over copyright concerns.⁴⁴ Decades of copyright precedent indicate that such a use, without permission, is not allowed.⁴⁵ Authors of copyrighted work allege that the tech companies are infringing on their copyrights.⁴⁶ Although AI companies often invoke the fair use doctrine to justify training their models with copyrighted works, creatives have pushed back forcefully.⁴⁷

Anthropic, a company sued for its unauthorized use of works, as of September 2025, settled its case for over 1.5 billion dollars, compensating authors for the use of over 500,000 books.⁴⁸ The case was the first substantive decision on how fair use applies to generative AI systems.⁴⁹ However, there have been split rulings on whether fair use applies to AI and the training it goes through. In Bartz v. Anthropic, originally, the judge agreed that the company’s use of the plaintiff’s books to train AI models was acceptable and fair use because the training was transformative and did not completely copy the books.⁵⁰ Nevertheless, the judge then ruled that using millions of pirated books was illegal and thus, needed to go to trial.⁵¹

C. Prominent Cases That Will Guide the Landscape

One of the original cases that demonstrated the issue of using technological advancements in a way that exploited copyrighted works was Google v. Authors Guild.⁵² In 2005, Google launched the Google Book Project, which involved bilateral agreements between Google and research libraries.⁵³ Google made digital scans of each book and indexed them all online without authorization from the copyrighted books’ authors.⁵⁴ Authors Guild, an association of authors of copyrighted books, sued Google for copyright infringement.⁵⁵ This case marked a turning point in determining how far media conglomerates would be able to stretch the fair use doctrine. Here, Google invoked fair use as a shield to protect it from massive potential liability.⁵⁶ The court determined that Google’s project was a “transformative” use that provided a public good by creating a new tool for research and discovery without significantly harming the market for the original books.⁵⁷ Ultimately, the court held that Google’s digitization of books was not copyright infringement and established a precedent that the use of copyrighted materials to train algorithms could be considered fair use if the use is transformative and doesn’t directly harm the market for the copyrighted work.⁵⁸

One decade later, Thomson Reuters v. Ross Intel was decided on a similar but more technologically advanced topic.⁵⁹ The case marked the first substantive decision on the fair use defense in an AI-related copyright dispute.⁶⁰ Thomson Reuters alleged that Ross directly infringed its copyrights by using Westlaw’s headnotes and Key Number Systems to train its AI legal search tool.⁶¹ More specifically, after being denied a license to use Westlaw’s legal research database for AI training purposes, Ross instead licensed 25,000 “Bulk Memos” from a third party to train its AI legal research tool.⁶² The court ultimately sided with Thomson Reuters, granting summary judgment after finding that the headnotes and key number system were original enough to be copyrightable and Ross’s action did not constitute fair use of the work.⁶³ Additionally, the court found that the purpose and character of the use was not transformative or different from Thomson Reuters’ use.⁶⁴

It is important to note that Ross specifically addressed the use of AI to create a competing product, which is different from addressing the generative AI claims that are present in the following cases. However, the analysis the court used is still helpful to determine what factors courts may hold in high regard in this area of the law.⁶⁵ Additionally, Rossis different from generative AI cases because the copyrighted works there were used to create competing tools, not competing content.⁶⁶ However, there is another lingering argument that, regardless of who created them, generative AI tools will undervalue authors by facilitating the creation of competing works, thereby diminishing their creative content in a way that should not be allowed.⁶⁷

Kadrey v. Meta is another case that ruled favorably for media conglomerates.⁶⁸ There, Meta developed and released an AI software that was trained by exposure to massive amounts of text from various sources.⁶⁹ This included authors like Kadrey, who filed a putative class action lawsuit against Meta, alleging copyright infringement for the unauthorized use of his works.⁷⁰ The case was decided on what the judge determined a fact-specific scenario.⁷¹ The court acknowledged that the transformative nature of using copyright works to train these AI models was influential in deciding whether the fair use exception is permitted; however, that alone does not guarantee fair use.⁷² The Kadrey court focused on the impact the use had on the market value, or potential market, of the original works.⁷³ If there is no evidence of market harm, then there is no injury that the court can resolve.⁷⁴ However, the court, ruling in favor of Meta, stated that it did not stand for the idea that Meta’s use of copyrighted materials to train its models was lawful.⁷⁵ Rather, in this case, the plaintiffs simply made the wrong arguments and did not develop a record in support of the right one.⁷⁶ The court also added that Meta’s acquisition of the books from unauthorized online repositories did not in itself preclude a finding of fair use for the training purpose, though the manner of acquisition could be relevant in other context.⁷⁷

The last case that details how courts may decide AI copyright cases is Bartz v. Anthropic, which ended in a landmark copyright settlement.⁷⁸ There, the court confronted the question of whether Anthropic’s use of copyrighted books to train its AI model was protected by the fair use doctrine.⁷⁹ A question that has frequently been avoided. The court concluded that using books to train a generative AI system is exceedingly transformative, likening it to how a human might read and then later draw upon a book’s themes and style to create new works.⁸⁰ The court emphasized that the AI’s output did not reproduce or closely mimic the plaintiff’s works, and the training process itself was fundamentally different from the original purpose of the books.⁸¹ However, the court differentiated this decision from Anthropic’s acquisition and retention of pirated books.⁸² While the use was transformative, the creation and maintenance of pirated works was not protected by fair use.⁸³ The judge made clear that pirating copies without paying for them is not transformative nor protected.⁸⁴

Although Anthropic was heard in the Northern District of California, it is likely to have a ripple effect on other AI and copyright lawsuits around the nation, as the case essentially stands for the proposition that legally acquired copyrighted works can be put into an LLM for AI training without legal ramifications.⁸⁵ While Anthropic settled and will owe authors over a billion dollars in damages, it is not the tidal wave win that creatives were hoping for.

Looking at the lawsuits being pursued, there are strong arguments on both sides. On the one hand, public interest promotes societal progress and supports the fair use of copyrighted works to do this. On the other hand, applying the doctrine so broadly undermines the protections that copyright law was intended to provide authors.

III. Discussion

As of November 2025, there are many active copyright lawsuits against AI companies that to continue grow.⁸⁶ The status of these lawsuits is highly volatile and varies significantly depending on the jurisdiction and the company involved.⁸⁷ Of these cases, three are currently on appeal, including Doe 1 v. GitHub, Thomson Reuters v. ROSS Intelligence, and Raw Story Media v. OpenAI.⁸⁸ Two cases, Bartz v. Anthropic and Vacker v. Eleven Labs, have settled.⁸⁹ In these cases, the companies argue that fair use allows them to use the information to train their models, while the authors hold tight to the idea that the fair use exception is limited in scope.⁹⁰ The court’s decisions have been split and have laid the groundwork for both sides to build stronger arguments to hopefully set a beneficial precedent.⁹¹ The cases that have already been decided or settled will likely be the most beneficial in shaping the potential landscape for future rulings.⁹² This Section analyzes some of the most important AI and copyright cases and uses past decisions to analyze ongoing cases. This Section also discusses fair use and how the once narrow doctrine is now being construed in a way that no longer provides authors the protection it once promised.

Examining previous cases, it is clear that courts continue to apply the fair use doctrine in ways that expand an exception once thought to be narrowly construed.⁹³ However, as the court detailed in Kadrey, there are many factors at play.⁹⁴ While Kadrey provided some clarification on factors courts look at when deciding newly presented questions about AI and copyright, it also unlocked more questions.⁹⁵ The court did not address whether or not a correctly made argument would allow for a ruling in favor of the plaintiffs, nor did it elaborate on why pirated materials could still be considered fair use, or ultimately answer whether Meta’s and many other media conglomerates’ activities are lawful.⁹⁶ These questions create risks and uncertainties for media companies training large AI models, risks that they often decide to take because the court has not explicitly deemed their behavior illegal.⁹⁷

The outcomes of the AI copyright cases differ because different courts analyze the factors and the weight of each differently.⁹⁸ Some courts view the market harm as the most influential factor in deciding if the fair use exception applies, while others emphasize the transformative nature of the copyrighted work in future use.⁹⁹ As of today, there is no clear-cut answer as to whether the fair use exception is a blanket exception that applies to all training of AI models. AI presents different challenges because the information fed to it is not transformed by a human.¹⁰⁰ Additionally, the work produced by artificial intelligence models can be transformed in many ways, but there are no ownership rights that the model claims, so the person using the model is likely the one who claims ownership of whatever is created.¹⁰¹

Anthropic is the most recently decided case and could very quickly become the new precedent around the country. It’s clear that courts are not abandoning the fair use doctrine; in fact, it is being strengthened through companies’ use of copyrighted works. This has led many to argue that authors are not being compensated justly for the use of their works.¹⁰² With how the courts are discussing the acquisition of copyrighted works, if the work is purchased legally and then fed into the system, there would be no copyright issue. What becomes difficult is the idea that AI produces work from someone else’s copyrighted work, which is then used by an individual as their own. However, up to this point, no plaintiff has been able to show their copyrighted work was substantially produced by the AI models. The courts have struggled to grapple with this because, without concrete evidence of injury credited to the AI, the fair use exception takes precedent when looking at all of the factors together.¹⁰³

For authors and creatives to potentially get a win in the courtroom, it is likely that more time and money is needed. The authors need to show that their work is being produced by AI word-for-word or significantly close to that. Without this, the Court will continually find that there is no market harm in a way that is relevant enough to ban companies from training these models on their work, as long as it is acquired legally and not pirated.

IV. Conclusion

In conclusion, authors have struggled to find a win in the world of AI, fair use, and copyright. Although the Anthropic lawsuit brought about a settlement that will pay out over a billion dollars in damages, the Court did not side with the plaintiffs on the fair use exception. The issue in Anthropic was the pirating of the works, not the use of them in AI training.

What may not be clear but subtly undertones the courts’ rulings is the want for technological advancement. Staying true to that is something that has clearly driven the courts’ decisions. If a court ruled against an AI company regarding the training of AI models and required developers to pay authors every time their work was used by an AI model, it would be impossible to track and much more costly than the technology already is. Judges and courts alike do not want to cause a slowdown in the development of AI. Thus, it’s clear that the public benefit argument and the lack of evidence regarding market harm dominate the decisions being made. The fair use exception is a perfect way to bolster the companies’ ability to use copyrighted works in an efficient way. However, it is unclear if this is what the fair use exception was truly intended for. Countering that, maybe the fair use exception is being used in the correct manner, and copyright law is just simply not strong enough to counter the argument in any way.

Regardless, it is unfortunate that authors and creatives lack a clear way to protect themselves and their works from potential copyright violations in the AI industry. As an author, it may not even be the idea that your work is being used by an AI model, but that it could potentially be used by someone else and passed off as their own. Authors put time and energy into creating pieces that are original, and copyright protection eases the thought that this could be taken away. Ironically, as time goes on, the fair use exception seems to be infringing on the power to claim copyright infringement in this industry.

Cover Photo by Rizki Ardia on Unsplash

Leah Luckett

References

1
Tanner Kohler, How AI Models Are Trained, Nielson Norman Grp. (May 2, 2025) https://www.nngroup.com/articles/ai-model-training/ [https://perma.cc/4B27-YREN].
2
Id.
3
Id.
4
Id.
5
Id.
6
Id.
7
Jorge Martinez Sanitago, The Hidden Cost of Complexity in AI Models, Medium (Mar. 17, 2025) https://medium.com/@jorgemswork/the-hidden-cost-of-complexity-in-ai-models-and-how-to-minimize-it-be8f7a868088 [https://perma.cc/DHL3-LQS2].
8
Id.
9
Id.
10
Ella Creamer, US Authors’ Copyright Lawsuits against OpenAI and Microsoft combined in New York with newspaper actions, The Guardian (Nov. 21, 2025) https://www.theguardian.com/books/2025/apr/04/us-authors-copyright-lawsuits-against-openai-and-microsoft-combined-in-new-york-with-newspaper-actions [https://perma.cc/H43S-PC4C].
11
Id.
12
Id.
13
What is Copyright?, U.S. Copyright Office, https://www.copyright.gov/what-is-copyright [https://perma.cc/MFG5-VP5A].
14
Id.
15
Id.
16
Matthew Sag, The Prehistory of Fair Use, 76 Brook L. Rev. 1371, 1378 (2011).
17
Lydia P. Loren, The Nature of Copyright: A Law of Users’ Rights, 90 Mich. L. Rev. 1615, 1624 (1992).
18
Id. at 1631.
19
Id. at 1627.
20
Id. at 1629.
21
Id. at 1632.
22
Id. at 1629.
23
Id. at 1625.
24
Sag, supra note 16, at 1371.
25
Loren, supra note 17, at 1632.
26
Id. at 1624.
27
Sag, supra note 16, at 1371.
28
Id. at 1372.
29
Authors Guild, Inc. v. Google Inc., 954 F. Supp. 2d 282, 294 (S.D.N.Y. 2013).
30
Authors Guild v. Google, Inc., 804 F.3d 202, 229 (2d Cir. 2015).
31
Kohler, supra note 1.
32
Id.
33
Id.
34
Id.
35
Id.
36
Id.
37
Id.
38
Id.
39
Id.
40
Id.
41
Id.
42
Id.
43
Id.
44
Katelyn Chedraoui, AI Has Sent Copyright Laws into Chaos. What You Need to Know About Your Rights Online, CNET (Nov. 11, 2025) https://www.cnet.com/tech/services-and-software/ai-has-sent-copyright-laws-into-chaos-what-you-need-to-know-about-your-rights-online/ [https://perma.cc/Y83K-SW65].
45
Id.
46
Id.
47
Id.
48
Chloe Veltman, Anthropic settles with authors in first-of-its-kind AI copyright infringement lawsuit, NPR (Sept. 5, 2025) https://www.npr.org/2025/09/05/nx-s1-5529404/anthropic-settlement-authors-copyright-ai [https://perma.cc/NYS5-JK8R].
49
Id.
50
Id.
51
Id.
52
Authors Guild, Inc. v. Google Inc., 954 F. Supp. 2d 282, 286 (S.D.N.Y. 2013).
53
Id. at 285.
54
Id. at 286.
55
Id. at 289.
56
Id. at 288.
57
Id. at 291.
58
Id. at 293.
59
Thomson Reuters Enter. Ctr. GmbH v. Ross Intel. Inc., 694 F. Supp. 3d 467, 478 (D. Del. 2023).
60
Id. at 475.
61
Id.
62
Id. at 476.
63
Thomson Reuters Enter. Ctr. GmbH v. Ross Intel. Inc., 765 F. Supp. 3d 382, 390 (D. Del. 2025).
64
Id. at 397.
65
Id.
66
Lisa T. Oratz, Fair Use Defense Failed in Thomson Reuters v. Ross, Jury Still Out for Generative AI, Perkins Coie, (March 27, 2025) https://perkinscoie.com/insights/update/fair-use-defense-failed-thomson-reuters-v-ross-jury-still-out-generative-ai [https://perma.cc/75VD-PA6F].
67
Id.
68
Kadrey v. Meta Platforms, Inc., No. 23-cv-03417-VC, 2023 LX 12120, at *5 (N.D. Cal. Nov. 20, 2023).
69
Id. at 2.
70
Id.
71
Id. at 1.
72
Id. at 2.
73
Id.
74
Id.
75
Id. at 5.
76
Id.
77
Id. at 1.
78
Bartz v. Anthropic PBC, No. C 24-05417, 2025 LX 456017, at *9 (N.D. Cal. Oct. 17, 2025).
79
Id. at 13.
80
Bartz v. PBC, 787 F. Supp. 3d 1007, 1019 (N.D. Cal. 2025).
81
Id. at 1021.
82
Id. at 1015.
83
Id. at 1019.
84
Id. at 1022.
85
David M. McIntosh, Regina Sam Penti, Yam Schaal & Maureen (Mo) Greason, A Tale of Three Cases: How Fair Use is Playing Out in AI Copyright Lawsuits, Ropes and Gray (July 7, 2025) https://www.ropesgray.com/en/insights/alerts/2025/07/a-tale-of-three-cases-how-fair-use-is-playing-out-in-ai-copyright-lawsuits [https://perma.cc/MQ5N-ZJMS].
86
AI Copyright suits Status Report, Chat GPT is Eating the World (Oct. 8, 2025) https://chatgptiseatingtheworld.com/2025/10/08/status-of-all-51-copyright-lawsuits-v-ai-oct-8-2025-no-more-decisions-on-fair-use-in-2025/ [https://perma.cc/H2B9-FQL7].
87
Id.
88
Id.
89
Id.
90
Id.
91
Id.
92
Sag, supra note 16, at 1411.
93
See Bartz v. Anthropic PBC, No. C 24-05417, 2025 LX 456017, at *26 (N.D. Cal. Oct. 17, 2025).
94
Kadrey v. Meta Platforms, Inc., No. 23-cv-03417-VC, 2023 LX 12120, at *5 (N.D. Cal. Nov. 20, 2023).
95
Id. at 3.
96
Id.
97
Id. at 6.
98
Id. at 4.
99
Id. at 6.
100
McIntosh, et al., supra note 84.
101
Id.
102
Id.
103
Id.