The fair use doctrine has long functioned as a critical mechanism for balancing the rights of original creators with the broader public interest in creativity, critique, and innovation. At its core, fair use permits the unlicensed use of copyrighted material in specific contexts—parody, commentary, research, criticism, and commentary—where the societal value of secondary use arguably outweighs the copyright holder’s exclusive rights. As the boundaries of creative expression continue to expand, particularly with the creation and utilization of digital tools and generative technologies, courts are increasingly faced with novel questions about the limits of transformation and the proper application of the fair use doctrine.
This article examines the evolving nature of fair use in the digital era, focusing on the central but often inconsistent role that the concept of transformation plays in judicial decisions. Drawing on two pivotal Supreme Court cases,Campbell v. Acuff-Rose Music, Inc. (1994), and Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith (2023), this analysis argues that although transformation remains a vital lens for understanding fair use, its current legal interpretation is unstable and ambiguous. The result is a chilling effect on artists and digital creators, whose works increasingly fall into gray areas of legality. This article concludes by proposing a more structured framework for determining transformative use that will better reflect the realities of 21st-century artistic production.
Under Section 107 of the U.S. Copyright Act, fair use is assessed using four non-exclusive factors: the purpose of character of the use; the nature of the copyrighted work; the amount and substantial; and the effect of the use upon the potential market. [1] The purpose of character of the use includes whether such use is commercial, for nonprofit educational purposes, and the amount of substantiality of the portion used in relation to the copyrighted work as a whole.
Although all four factors are theoretically weighed equally, courts have placed considerable emphasis on whether the use is “transformative.” A work is said to be transformative if it furthers a purpose or presents a different character, differentiating itself from the original with a new expression, meaning, or message. However, as subsequent cases have demonstrated, the standard for determining what qualifies as transformative remains highly subjective and inconsistently applied.
The concept of transformation first gained prominence in Campbell v. Acuff-Rose Music, Inc. (1994), where the Supreme Court upheld the fair use defense of a commercial parody by 2 Live Crew. The group’s version of Roy Orbison’s “Oh, Pretty Woman” was deemed transformative because it offered social commentary through parody. [2]This case set an important precedent: commercial works could be fair use if they significantly transformed the original. It also introduced the notion that transformation is not solely about style or form but message and purpose as well. Over time, courts increasingly used this test as the centerpiece of their fair use analyses. Yet what began as a clarifying lens has since become a murky legal standard.
The ambiguity surrounding transformation came to the forefront again in Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith (2023). The dispute concerned Warhol’s use of a 1981 photograph of Prince taken by photographer Lynn Goldsmith. Warhol had created a series of stylized portraits based on the photograph, and after his death, the Andy Warhol Foundation licensed one of the images to Vanity Fair without Goldsmith’s permission. Goldsmith sued, arguing that Warhol’s work infringed her copyright.
The Supreme Court, in a 7-2 decision, sided with Goldsmith, holding that Warhol’s image was not sufficiently transformative to qualify for fair use. The majority emphasized that both the photograph and Warhol’s work served the same commercial purpose: to illustrate a magazine story about Prince. Thus, despite Warhol’s stylistic changes and artistic reputation, the Court concluded that the work lacked the necessary shift in meaning or message to be “transformative”. [3]
The decision has far-reaching implications. By focusing heavily on market substitution and downplaying the significance of aesthetic or interpretive transformation, the Court narrowed down the transformative use test. The result is an environment in which artistic reinterpretation—especially in visual and digital media—faces heightened legal risk. Justice Elena Kagan noted, “creativity relies upon the borrowing of works that came before” [4]
The contrasting outcomes in Campbell and Warhol highlights a growing inconsistency in the application of the fair use doctrine. In Campbell, the Court celebrated parody as a transformative use deserving of protection, even when its enforcement was commercially motivated. In Warhol, similar commercial use was deemed disqualifying, even though Warhol’s artwork arguably introduced new artistic insights and commentary.This disparity reflects deeper tensions in how courts evaluate artistic transformation. Should transformation hinge on a work’s purpose, style, or meaning? Or should courts prioritize whether the new work competes with the original in the same market? Without clearer answers, artists are left navigating a legal minefield, unsure whether their derivative works fall within the bounds of fair use.
If transformation is not clearly defined, this challenge will continue to arise in new edge cases—particularly in the domain of generative work. Generative AI introduces an entirely new dimension to fair use analysis. These systems often rely on vast datasets scraped from the internet, which may include copyrighted material. While the models do not reproduce content verbatim, they are trained on it, raising questions about whether the training process itself constitutes a form of infringement. What happens when an AI model generates images “in the style of” an artist or trains on copyrighted content? Is the output transformative? Or is it an unauthorized derivative imitating the work of other artists?
Many legal scholars argue that machine learning (ML) systems should generally be allowed to use copyrighted materials for training under the fair use doctrine. As Dan Cahoy notes, the first factor of fair use “favors…if the work is ‘transformed’ into something that is different in terms of its message or use from the original,” and this might include “fresh information, artistic elements, insights, or viewpoints beyond simply copying the original.” Mark Lemley, professor at Stanford Law, furthers this point by distinguishing between the protected expression of a work and the unprotected elements—such as facts, ideas, and structure—that AI models seek to learn. [5] Lemley writes that ML systems copy, not to exploit creative expression, but to “get access to the uncopyrightable parts of the work… the ideas, facts, and linguistic structure.” [6] In this view, the training of AI models may qualify as a type of ‘fair learning’ even if some fair use factors (like the amount used)might otherwise weigh against it.
Similarly, the Library Copyright Alliance (LCA) supports the idea that training AI on copyrighted content is generally transformative and thus fair use. In their submission to the U.S. Copyright Office, the LCA emphasizes that “the ingestion of copyrighted works to create large language models or other AI training databases generally is a fair use.” [7] The Alliance draws on precedents such as Authors Guild v. Google and Authors Guild v. HathiTrust, where courts upheld mass digitization from non-expressive machine learning as transformative and fair. These cases, while not generative in nature, confirm that the fair use doctrine is flexible enough to accommodate learning-based technologies.
Limiting AI training data to only public domain materials could have chilling effects on research and innovation. As scholars from UC Berkeley Library write, requiring licenses for AI training would “hamper researchers’ ability to interrogate modern in-copyright materials” and skew the knowledge base toward outdated or non-representative works. [8] If courts interpret fair use too narrowly, they risk entrenching monopolies over cultural data and stifling public access to tools that could democratize knowledge, art, and expression.
Fair use remains one of the most vital, yet most contested, aspects of copyright law. As artistic practices evolve, especially in the digital and generative domains, the current legal framework constructing the transformative use testhas proven insufficiently clear and consistent. While landmark decisions like Campbell opened the door for expansive interpretations of fair use, recent rulings such as Warhol suggest a judicial retrenchment, raising concerns about the future of artistic freedom. To protect both original authors and derivative creators, courts and lawmakers must rethink and refine the boundaries of fair use in the modern era.
Works Cited
[1] 17 U.S. Code § 107
[2] Campbell v. Acuff-Rose Music, Inc.
[3] Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith
[4] Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith
[5] “AI Training Data Dilemma: Legal Experts Argue For ‘Fair Use.’”
[6] “AI Training Data Dilemma: Legal Experts Argue For ‘Fair Use.’”
[7] “Training Generative AI Models on Copyrighted Works Is Fair Use.”
[8] “Training Generative AI Models on Copyrighted Works Is Fair Use.”
