Anthropic has secured a significant legal victory in a landmark copyright case, with U.S. District Judge William Alsup ruling that the company’s use of legally purchased books to train its AI models, including the Claude chatbot, constitutes fair use under U.S. copyright law. This marks the first time a federal court has explicitly endorsed the practice of training large language models on copyrighted works without permission, provided the materials were lawfully acquired. The decision is a major win for the AI industry, potentially setting a precedent for how courts may handle similar cases involving generative AI and intellectual property. The lawsuit was filed by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, who accused Anthropic of “strip-mining” their copyrighted works to train its AI systems. They argued that the company’s actions amounted to large-scale theft and undermined the economic value of creative labor. However, Judge Alsup rejected this argument, emphasizing that the training process is transformative. He compared it to a student reading countless books to become a writer—learning from the material without copying or replacing it. “Like any reader aspiring to be a writer, Anthropic’s LLMs trained upon works not to race ahead and replicate or supplant them—but to turn a hard corner and create something different,” Alsup wrote. The judge concluded that training AI models on digitized versions of books purchased in physical form is a fair use because the resulting AI outputs are not direct copies of the original works. Instead, they represent a new form of expression derived from broad patterns, grammar, and style across thousands of texts. Alsup described the technology as “among the most transformative many of us will see in our lifetimes,” underscoring its potential to advance scientific progress and creativity. However, the ruling is not a complete victory for Anthropic. Judge Alsup also found that the company’s practice of downloading millions of pirated books from the internet—many of which were obtained from unauthorized websites—was not protected by fair use. The court will hold a separate trial to determine damages related to this conduct. Alsup stressed that even if the pirated books were later used for training, their initial acquisition through illegal means cannot be justified under copyright law. He noted that buying a copy after stealing it does not erase the original infringement. Anthropic acknowledged the partial win and expressed satisfaction with the court’s recognition of the transformative nature of AI training. A company spokesperson said the decision aligns with the purpose of copyright: to encourage innovation and progress. The company also noted that it has since shifted its approach, hiring a former Google Books executive and moving toward more lawful sourcing of content. The ruling comes amid growing legal pressure on AI companies. The New York Times, for example, has sued OpenAI and Microsoft over the use of its articles to train AI models. Other publishers and creators are pursuing similar claims, while some are negotiating licensing deals with AI firms. This case highlights a critical divide: the AI industry’s need for vast data to innovate versus creators’ rights to control and profit from their work. Though this decision is binding only in the Northern District of California, it could influence future rulings nationwide. The outcome may also prompt legislative action to modernize copyright law for the age of AI. For now, the ruling provides a legal foothold for AI companies to argue that training on copyrighted material, when done responsibly and with lawful acquisition, can be a legitimate, transformative use under the law.