The Undergraduate Law Review at UC San Diego

This Article explores a discrete yet consequential driver of AI-generated copyright infringement: overfitting. By analyzing overfitting as a statistical phenomenon, it offers a legal framework for understanding how this technical failure can lead machine learning models to reproduce protected works nearly identically. Through the lens of the Italian Plumber Problem, this article critiques the limitations of the fair use doctrine when applied to generative AI outputs, identifies regulatory and policy shortcomings, and argues for an updated doctrinal toolkit. Finally, it draws a conceptual parallel between algorithmic overfitting and judicial reasoning, positing that courts may fall prey to overfitting by relying on fact-specific precedent in the absence of guiding rules for novel technologies. Using interdisciplinary analysis, this article calls for a recalibration of legal frameworks to accommodate the realities of machine learning and protect both innovation and authorship in the digital age.