Artificial intelligence raises various novel legal questions about data privacy, bias in outputs, cybersecurity, and ethics, among other topics. In 2023, however, intellectual property concerns associated with artificial intelligence took center stage. In the last year, at least twelve lawsuits have been brought against generative AI companies raising copyright infringement and related claims.
Almost all of these are putative class actions brought by authors or visual artists, seeking money damages on behalf of some defined class of people whose copyrighted works were used to train the defendant’s AI model. In addition, some companies (most notably Getty Images, The New York Times, and a set of music publishers) have filed non-class-action lawsuits based purely on the use of their own copyrighted works to train AI models.
While the parties and legal theories are often similar, plaintiffs have had varying degrees of success. For instance, in an earlier-filed case involving the alleged use of Westlaw content to train a legal AI tool, the district court issued a decision in September that allowed Thomson Reuters’ copyright claims to proceed to trial and held that the defendant’s fair use defense was more appropriate for a jury. By contrast, courts have so far been more skeptical of certain claims in the more recently filed class action lawsuits against generative AI companies. For instance, multiple courts have rejected the theory that an AI model trained on copyrighted works is an infringing derivative work, or that anything created with that model is an infringing derivative work, even if the outputs of that model are not “substantially similar” to the works used as training data. Even in these cases, however, defendants’ motions to dismiss in these cases have often not challenged the plaintiffs’ infringement allegations that their works were copied without authorization to be used as training data. This is likely because defendants recognize that this implicates questions about fair use that courts typically are unwilling to resolve early in a case by way of a motion-to-dismiss.
The facts on which complaints are based have also varied in interesting ways. A recently amended complaint in Kadrey v. Meta Platforms, Inc., for example, includes Discord chat logs from a Meta-affiliated researcher stating that he had discussed with Meta’s legal department whether use of certain ebook files to train AI would be “legally okay” and the legal department had raised concerns about certain uses1 . The plaintiffs rely on that allegation to argue that Meta engaged in willful infringement. Different plaintiffs have also had different degrees of success in arguing that AI tools have generated “substantially similar” outputs, with some able to point to specific examples of AI tools appearing to copy from their copyrighted works2 . Future court decisions may focus on these differences in allegations.
While this wave of litigation is only the beginning as AI becomes more prominent, these cases may shape the U.S. legal landscape that will govern AI in a variety of areas.
Footnotes
[1] Am. Compl. ¶¶ 55–63, Kadrey v. Meta Platforms, Inc., No. 323-cv-3417 (N.D. Cal. Dec. 11, 2023), ECF No. 64.
[2] Compl. ¶¶ 66–69, Concord Music Grp., Inc. v. Anthropic PBC, No. 23-cv-1092 (M.D. Tenn. Oct. 18, 2023), ECF No. 1.
[View source.]
