For months, both the artist and IP legal communities eagerly awaited the US Supreme Court’s decision in the case of Andy Warhol Foundation for the Visual Arts, Inc. v. Lynn Goldsmith et al. (Warhol). Many hoped the decision might provide some guidance on copyright issues raised by the proliferation of generative AI tools. Like most machine learning models, generative AI systems are typically trained using large datasets, including copyrighted materials, to identify and replicate patterns. AI developers have long taken the view that these training practices fall under the “fair use” doctrine, submitting that their purpose and character is highly transformative, relying on decisions that have held that the re-purposing of protected materials for the creation of new tools constitutes fair use (see Kelly v. Arriba Soft Corp., 366 F.3d 811 (9th Cir. 2002); Authors Guild v. Google, Inc., 804 F.3d 202 (2nd Cir. 2015); Google LLC v. Oracle Am., Inc., 141 S. Ct. 1183 (2021)).
Some recent cases have raised the question whether this debate is indeed settled. The Court’s 7-2 decision finding Warhol did not make fair use of Goldsmith’s photographs does not address these issues directly. But by focusing specifically on the character of the narrowly defined secondary use at issue, the Court could be suggesting that the fair use defense for generative AI tools will have to be determined on a case-by-case basis.
The Warhol Case
In the 1980s, photographer Lynn Goldsmith granted a limited license to Vanity Fair to use her portrait of the artist Prince as an artist’s reference to create one illustration to accompany an article about Prince. Under this license, Andy Warhol created a purple silkscreen portrait of Prince, which appeared alongside an article about the musician in Vanity Fair’s November 1984 issue. Unbeknownst to Goldsmith at the time, Andy Warhol went on to create a series of 16 works known as the “Prince Series”, all based on Goldsmith’s photograph. When Vanity Fair used one of the works in the series, an orange silkscreen portrait of Prince on the cover of its April 2016 issue, commemorating the late musician, Goldsmith brought suit claiming copyright infringement. The Warhol Foundation sought to use the fair use defense. Recently, the US Supreme Court affirmed the Second Circuit’s decision in favor of Goldsmith based on a finding that the first fair use factor, the purpose and character of the use, weighed against the Warhol Foundation.
To reach this conclusion, the Court clarified that the inquiry into the character and purpose of the use does not stop at determining whether the secondary use has a different message, meaning, or purpose. Rather, the purpose and character of the secondary use itself must not be aimed at supplanting the original work. The Court narrowly focused on the specific circumstances on the case: Because Goldsmith’s original photo and Warhol’s unlicensed secondary use both had the purpose of accompanying magazine stories about Prince, the Court found them to share substantially the same commercial purpose. Informed by these circumstances, the Court formulated the general rule that “[i]f an original work and a secondary use share the same or highly similar purposes, and the secondary use is of a commercial nature, the first factor is likely to weigh against fair use, absent some other justification for copying.”
Using Generative AI Tools in an Uncertain Legal Environment
The commercial potential of generative AI tools is undisputed. As recently illustrated by the viral success of an AI-generated song mimicking the voice and style of the rapper Drake, AI-generated outputs also have a real capability to rival new works by creators of the works on which these models were trained. But this does not necessarily close the door to fair use defenses for AI tools. The outcome of the analysis of the first fair-use-factor may vary across generative AI tools and depending on whether the alleged infringement is in the training, the prompts, or the output, as the character and purpose of the use at each stage of generative AI development and use may differ. Ultimately, what the Warhol decision means for developers and users of generative AI tools is that the availability of a fair use defense continues to require a careful case-by-case analysis.
While case law on generative AI continues to evolve, players in this ecosystem, whether it be the content creators, content licensors or AI developers, are already taking action to protect their rights and shield their liability. Commercially, a number of platforms have introduced separate license agreements to use their data for commercial purposes, including the training of generative AI programs. If the courts find the training of generative AI tools on protected materials to be fair use, such contractual arrangements may be federally preempted.
In the meantime, however, contractual terms, including terms of use, can provide some clarity where the law is still unsettled as well as an opportunity for creators and AI developers to balance their respective interests and reduce risks in an environment marked by regulatory and legal uncertainty and rapid technological development. In this vein, it is worth noting that in Warhol, Goldsmith had granted a limited license to Vanity Fair to use her photograph to create an illustration to accompany an article about Prince. Ultimately, it may have been the violation of this limited license by the Warhol Foundation’s licensing of another use of Goldsmith’s work to the same end that prompted this litigation.
