top of page

Fair Use and Generative AI

Updated: May 1

generative ai and copyright

As noted in our recent post and Case Tracker, there are numerous lawsuits challenging the use of copyrighted content in generative AI. The plaintiffs assert that AI models infringe by using copyrighted material in their training process, outputs, or both. AI providers assert that their use of copyrighted content is permitted by the fair use doctrine. The question of “fair use” may be outcome determinative in these cases. Read on for a refresher on fair use and key precedents likely to be influential in the pending cases. (Use our links to the Copyright Office’s excellent Fair Use Index for more detailed case summaries and citations.)

Fair Use

The fair use doctrine seeks to strike a balance between fostering the creativity of authors and promoting the public's benefit from accessing expressive materials. As previously explained, courts consider four factors set forth in Section 107 of the Copyright Act to determine whether a particular use constitutes fair use: (1) the purpose and character of the use; (2) the nature of the copyrighted work; (3) the amount and substantiality of the portion used; and (4) the effect of the use upon the potential market for the copyrighted work. In recent years, courts have increasingly relied on whether a new use is “transformative” in reaching decisions.

Reverse Engineering and Interoperability


Reverse engineering, such as deconstructing a product to understand its components, can be fair use. For example, in Sega v. Accolade, the court held that reverse engineering source code to develop a video game that would run on Sega’s game console was fair use. Similarly, in Sony v. Connectix, the court found that intermediate copying Sony’s BIOS to develop a virtual game emulator that allowed users to play PlayStation games on their computers was fair use.

Thumbnails, Digitization and Snippets

In some cases, copying to provide information about a work, or make it more easily found, has been held to constitute fair use. For example, in Perfect 10 v., the court held that a search engine’s use of low-resolution, “thumbnail” versions of images, with links to source information, was fair use. In Authors Guild, Inc. v. HathiTrust, the court held that creating digital copies of library books to create a full text searchable database was fair use because it was transformative and did not harm any existing or potential markets for the books. Similarly, in Authors Guild v. Google, the court held that scanning and digitizing 20 million library books to make them searchable and displaying short excerpts, or snippets, was transformative fair use that augmented public knowledge about the books without substituting for them.


The Supreme Court found fair use in Oracle v. Google based on Google’s copying of 11,500 lines of Java declaring code and its use of the Java organizational structure to enable programmers familiar with Java to more easily write programs for its Android operating system.

It’s Not Fair (Use)!

There are thousands of fair use decisions, and many of them do not reach a finding of fair use. The Supreme Court’s most recent fair use decision, in Andy Warhol Foundation v. Goldsmith, found that the first fair use factor (nature and purpose of the use) favored Goldsmith and weighed against a finding of fair use where a magazine published a Warhol silkscreen created using Goldsmith’s photo without permission. As explained in a previous post, the decision relied heavily on the fact that both the original photo and the silkscreen had been used to illustrate articles about Prince. This ruling may lead courts to reign in their reliance on transformativeness as a shortcut for fair use analysis.

In Harper & Row Publishers v. Nation Enters., the Supreme Court held a news magazine’s use of an unauthorized copy of a forthcoming presidential biography was not fair use because the excerpts went to the heart of the book and harmed the market for licensing use of excerpts. In Micro Star v. FormGen, the court held that use of characters and images from a game within a compilation of user-generated levels for the game was not fair use because it made use of the game’s original creative elements without sufficient transformation. In Associated Press v. Meltwater, the court held that reproduction of articles by a news clipping service was not transformative fair use because it did not add significant value or alteration.

The Road Ahead

Courts deciding the pending AI cases are likely to wrestle with the doctrine of fair use, and the meaning of transformation in light of the Supreme Court’s decision in Andy Warhol Foundation. We anticipate some guidance from rulings on motions to dismiss; later summary judgment rulings should provide deeper analysis. The rulings will be case specific and the body of caselaw is likely to evolve gradually. Nevertheless, these decisions may have a dramatic impact on AI and the market for online content. Watch this space for updates.

nancy mertzel - partner of mertzel law pllc

Nancy J. Mertzel

Mertzel Law PLLC

1204 Broadway, 4th Floor, New York, NY, 10001

25 Pompton Avenue, Suite 101, Verona, NJ 07044

(646) 965-6900

Recent Posts

See All


bottom of page