Algorithmic Authoring and the Fair Use Doctrine: An Empirical Analysis of Judicial Interpretations regarding Generative AI Training Data in US Federal Courts

This study presents a systematic empirical analysis of how United States federal courts are interpreting and applying the fair use doctrine under 17 U.S.C. § 107 to the unauthorized use of copyrighted works as training data for generative artificial intelligence (AI) models. Analyzing a hand-coded dataset of 38 federal copyright lawsuits filed between January 2022 and May 2024, we examine the procedural trajectories, pleading standards, and judicial reasoning concerning the four statutory fair use factors. Our empirical findings reveal that while plaintiffs struggle to survive motions to dismiss regarding claims of derivative liability and output-based infringement, courts consistently preserve direct infringement claims centered on the unauthorized copying of training data for discovery. Furthermore, our qualitative analysis of early judicial opinions indicates a structural shift in the evaluation of 'transformativeness' under the first fair use factor, heavily influenced by the Supreme Court's decision in Andy Warhol Foundation v. Goldsmith. This article charts these judicial trends, offering a predictive framework for how fair use defenses will be adjudicated at the summary judgment stage, and discusses the broader implications for the balance between technological innovation and creators' rights.