The legality of training large AI models on copyrighted books remains unsettled as courts consider whether such use qualifies as fair use. Most major AI developers, including OpenAI, Google, Anthropic, and others, train their models on massive datasets that comprise published books and other copyrighted material without obtaining permission from authors. In 2025, Judge William Alsup ruled that Anthropic’s training of its AI models was lawful, though he ordered a $1.5 billion settlement for pirating books from illegal shadow libraries. Attorney Cathy Gellis said the decision treats model training like reading rather than copying, a view that benefits AI companies. She noted that copyright law hinges on copying, not on using or experiencing a work. Jason Henderson, a senior attorney, said the law has not kept pace with AI and that judges are divided on whether training that directly competes with original works infringes copyright. The Thomson Reuters v. Ross Intelligence case found that training on Reuters’ content for a competing AI legal platform was not fair use because it lacked transformative purpose. While some argue that chatbots compete with authors by generating synthetic works, that argument has not yet prevailed in court. Ongoing litigation means the legal framework for AI training will continue to evolve, and future rulings will determine the boundaries of permissible use under existing copyright law.