The rise of generative artificial intelligence (“AI”) has brought questions regarding the permissibility of using copyrighted materials in training generative AI systems around the globe. This article examines whether such use constitutes infringement under the American fair use doctrine, drawing insights from the regulatory approaches of the European Union (EU) and Japan.
Ultimately, this article advocates for United States courts to recognize the permissibility of using copyrighted materials in training generative AI models, rejecting the EU’s rights-holder-centric approach, and going beyond Japan’s limited-permissibility solution. Through a detailed four-factor fair use analysis, this article demonstrates that training generative AI on copyrighted materials weighs in favor of fair use. The transformative nature of this training fundamentally alters the purpose and character of the copyrighted works, and the societal benefits and technological progress enabled by this training outweighs the potential market impacts on rights holders. Furthermore, in evaluating the EU’s rights-holder-centric regulations and Japan’s “non-enjoyment” provision, which restricts generative AI training based on the intended output of the model, this article proposes that United States courts adopt a modified version of Japan’s non-enjoyment provision in their fair use analysis. This adoption would allow United States courts to accommodate rapid technology development and properly distinguish the permissibility of training generative AI systems from the permissibility of these systems’ output. This article concludes that American copyright law should foster technological growth by finding the use of copyrighted materials in training generative AI systems to be fair use.