The fair use doctrine in copyright law has been a hot topic in recent years in response to the rapid rise and development of generative artificial intelligence (AI) systems. As the adoption of generative AI accelerates and expands, lawyers, judges, artists, and consumers have been wondering what the United States Copyright Office (USCO) has to say about generative AI and the fair use doctrine.
The time has finally come for an answer. In a pre-publication version of its final report of a three-part series on copyright law as it relates to generative AI, the USCO states that some methods of training and purposes of using generative AI systems would not qualify as a fair use, while others would.
The USCO analyzed all four fair use factors, but the first and fourth factors dominated the analysis. The USCO focused heavily on the first factor, specifically the ultimate purpose of the use, highlighting the "transformativeness" analysis. On the one hand, the USCO reasoned that training an AI system to deploy that system for research purposes is likely transformative, while training an AI system on images from a popular animated series to deploy that system to generate images from that series is likely not. Most cases, the USCO opined, will fall somewhere in between. Interestingly, the USCO observed that if an AI system employs appropriate and effective guardrails preventing users from generating excerpts of copyrighted works, "the system will be less capable of fulfilling the purpose of the original works, and their use in training may be more transformative."
The USCO also zeroed in on the fourth factor, "the effect of the use upon the potential market for or value of the copyrighted work." It addressed how using copyrighted works to train generative AI systems may affect the markets for or values of the works, "including through lost sales, market dilution, and lost licensing opportunities." For lost sales, the USCO provided, as an example, someone using pirated collections to build a library of training data. For market dilution, the USCO took a broad view, discussing how generative AI systems create more competition for human artists and dilute their royalties and how allowing those systems to imitate an artist's or writer's style could harm that artist's or writer's market. And for lost licensing opportunities, the USCO opined that where licensing markets already exist to meet the training needs of an AI system, "unlicensed uses will be disfavored under the fourth factor," but if those markets do not exist, "there will be no functioning market to harm and the fourth factor may favor fair use."
The USCO also discussed various licensing models for procuring AI training data, including compulsory licensing, extended collective licensing, and the opt-out model. The USCO observed that voluntary direct and collective licensing arrangements have emerged, but the office did not endorse any particular model specifically. Adopting an objective, wait-and-see approach, the USCO instead recommends allowing the licensing markets to develop organically without compelling licenses or allowing government intervention.
The report has major implications for companies using or developing a generative AI system. However, the USCO's report is designated as a "pre-publication" report, and therefore, it is subject to change. Importantly, the USCO provides guidance only; it is not a law or a court decision with binding effect. At the same time, however, courts and policymakers will likely look to the USCO for help crafting opinions and laws relating to generative AI, and the more than 100 pages of this report contain helpful information and analyses relating to fair use. To help determine whether fair use applies to the use of a generative AI system, contact Taft's generative AI team.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
