When Market Disruption Is Not Market Harm: AI Training, Causation, and the Boundaries of Fair Use

Generative artificial intelligence systems are trained on vast quantities of data that include copyrighted works, prompting litigation over whether such training constitutes infringement or fair use. Much of this debate centers on claims of market harm, with copyright holders arguing that AI-generated outputs displace human creators and threaten creative industries. This paper offers a structured analysis of that dispute by isolating AI training as the relevant use for fair use purposes and examining how the fourth fair use factor applies to claims of widespread economic disruption.

The paper argues that AI training is plausibly transformative under the first fair use factor and that the copying at issue occurs upstream from the generation of competing outputs. It then traces the causal connection between training and creative market disruption, showing that training can satisfy traditional notions of but-for and proximate cause. Despite this causal linkage, the paper concludes that the harm most clearly associated with generative AI is directed at future human creativity rather than at the markets for identifiable copyrighted works.

As a result, the disruption-however real-may fall outside the category of market harm the fourth fair use factor is designed to recognize. Finally, the paper suggests that where copyright doctrine reaches its limits, alternative legal frameworks may offer more appropriate tools. In particular, it identifies unjust enrichment as a potential post-copyright pathway focused on value capture rather than market substitution, providing a possible response to economic effects that fair use doctrine cannot readily address.