Reassessing the Fair Use Doctrine in the Age of AI: Evaluating Bad Faith and Market Harm through Bartz v. Anthropic and Kadrey et al. v. Meta Platforms, Inc.

Abstract

This paper examines how the fair use doctrine applies to large language model (LLM) training by analyzing Bartz v. Anthropic and Kadrey v. Meta Platforms, Inc. LLMs such as “Claude’’ and “Llama’’, generate text from large datasets, raising novel copyright questions. In Bartz, lawfully acquired works were used fairly, while pirated copies weighed against fair use. In Kadrey v. Meta Platforms, Inc., Meta’s use of unauthorized sources was deemed transformative and fair, highlighting inconsistent treatment of bad faith. Regarding market harm, courts focused on direct substitution but largely ignored broader effects such as market dilution or long-term impacts on creative markets. The paper agrees that fair use doctrine must evolve with technology, proposing a fifth factor focused on the legality of AI training sources and greater attention to market dilution and long-term economic harm to creators.