The U.S. Supreme Court’s recent decision in Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith is unlikely to shed much light on whether the use of copyrighted material in artificial intelligence (AI) content will lead to liability. The Court’s decision mandates that courts look to the “specific use” of the copyrighted material at issue when evaluating fair use under the Copyright Act. So, what specific factors should AI developers and users consider when using copyrighted content in the AI space post-Warhol?

The Copyright Act and Generative AI

Under the Copyright Act, copyright holders have the exclusive right to reproduce their work, prepare derivative works, distribute copies of the work, perform the work, and display the work publicly. In developing an AI system, programmers and companies can violate exclusive rights of copyright holders at two distinct points:

• By using copyrighted material as an input to teach the AI software
• By creating an unauthorized derivative output of the copyrighted work because of the AI application. The distinctions between inputs and outputs involving this space are detailed here and here.

For example, the organizers of a summer camp want to use a famous pop song in an ad, but they don’t believe they’ll be able to get permission. Instead, they digitally copy the song as short-term background music and create a similar new song (that doesn’t infringe on the original), using an AI music generator. Even if the short-term version is not released, and the new version is not substantially similar to the original, the original internal copy still constitutes intermediate infringement, because it has reproduced the original song.

Although it is unlikely that the summer camp would be sued for making an interim copy that is never publicly released, it is important to note that whether an interim copy is public or private, commercial or noncommercial, it is still considered a copy.

Copyright law decisions involving computer programming and interoperability have allowed for video game and software companies to “reverse engineer” code from copyrighted programs as an interim step to creating a non-infringing final work. For example, in Sega Enterprises v. Accolade, video game maker Accolade copied Sega’s games to access functional code necessary to create games compatible with Sega’s Genesis gaming console. The Ninth Circuit reversed a lower court’s decision in favor of Sega and held that while the Copyright Act “unambiguously encompasses and proscribes” intermediate copying, Accolade’s conduct was permissible under the fair use doctrine. The court reasoned that Accolade’s intermediate copying constituted fair use because it was a necessary interim step to acquire the compatibility code (which is not copyrightable), and it served the public interest by increasing access to independently designed video games.

Therefore, if one of the exclusive rights of a copyright holder has been violated, liability under the Copyright Act can still be avoided if the unauthorized use of the copyrighted material constitutes “fair use.” Use of the copyrighted material “for purposes such as criticism, comment, news reporting, teaching (including multiple copies for classroom use), scholarship, or research” is considered fair use. Additionally, the Copyright Act provides four factors that are evaluated when determining whether use of a work is fair use.

Each fair use factor is balanced by courts when undertaking the fair use analysis; however, the Supreme Court’s decision in Warhol focused on the first fair use factor: the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes. In Warhol, the Court evaluated the purpose and character of the Andy Warhol Foundation’s licensing of Orange Prince, which was derived from Lynn Goldsmith’s copyrighted photograph of musician Prince Rogers Nelson. Applying the recent guidance of Warhol, and the use of a generative AI music creator in the above example, the remainder of this article will focus on whether use of copyrighted content in a generative AI output can constitute fair use, as delineated by the purpose and character fair use factor.

Can AI Use of Copyrighted Material Constitute Fair Use?

While courts have yet to apply the Supreme Court’s guidance in Warhol, whether the purpose and character of an AI output favors fair use will be case specific as courts have to evaluate the specific facts surrounding the use of the copyrighted material at hand. According to the decision, courts should consider “the reasons for, and nature of, the copier’s use of an original work. The ‘central’ question it asks is ‘whether the new work merely ‘supersede[s] the objects’ of the original creation . . . (‘supplanting’ the original), or instead adds something new, with a further purpose or different character.” Where instances of potential fair use are not “clear cut,” courts must evaluate the degree to which the copier’s work has a different purpose or character in the specific use that is being reviewed.

In the instance of the Warhol case, the Court evaluated the Warhol Foundation’s use of the derivative Orange Prince in the context of magazine licensing, an area in which Goldsmith had used her photographic work for the same purpose. In finding that the purpose and character were largely unchanged, the Court found against fair use in the case at hand but left open the door for fair use to be found in other instances by distinguishing parody and satire from the Warhol Foundation’s use of Goldsmith’s photograph.

The Warhol decision also further clarifies how courts should evaluate where a copier’s use is commercial as opposed to non-commercial, in tandem with the purpose and character of the work. “The commercial nature of the [copier’s] use is not dispositive[,]”but instead “it is to be weighed against the degree to which the use has a further purpose or different character.”

Accordingly, whether an AI output derived from copyrighted material can make the purpose and character of the use different enough from the original work to favor fair use will be fact specific. For example, consider again the hypothetical summer camp submitting a famous pop song to an AI music generator. When evaluating the first fair use factor in that case, a court would likely try to determine whether the song in the summer camp ad merely supplants the original or adds something new that gives it a different purpose or character. In that instance, a court may determine that the purpose of the new song favors fair use in that it is intended to be marketed to children, and that the original song was likely not used (or intended) for the purposes of a commercial jingle. Likewise, if the new AI-generated song criticizes or parodies the original, the purpose and character of the AI work can favor a finding of fair use.

If the AI output merely creates a work piggybacking off elements of the original work, fair use is unlikely to be found, particularly where the AI use is commercial and potentially usurps demand for the original copyrighted work. Some AI applications currently charge commercial subscriptions to access their features, which can weigh against a finding of fair use in circumstances where an AI output does not sufficiently transform the original copyrighted work.