Using Copyrighted Works to Train Artificial Intelligence: A New Stage in the Development of the Fair use Doctrine

Abstract

The article analyzes the landmark order of the US District Court (Northern District of California) in the case of Bartz et al. v. Anthropic PBC (2025), which became the first precedent for a comprehensive assessment of the application of the fair use doctrine in relation to the practice of training large language models on copyrighted materials. Based on a detailed analysis of the court's reasoning, the author demonstrates where the line of legality for using data lies: using copyrighted objects to train artificial intelligence and to create a digital library by digitizing legally acquired books was recognized as fair use, while using pirated copies to create such a library was considered unfair use. The purpose of this article is to analyze the key legal positions formulated by the court in this order, their significance for understanding the evolution of the fair use doctrine in the digital age and the practical consequences for artificial intelligence developers and authors of works.