AI Training Deemed Fair Use, But Pirated Book Storage Sparks Trial
A U.S. judge has ruled that Anthropic’s use of copyrighted books to train its artificial intelligence system qualifies as fair use, dealing a significant blow to authors challenging how AI companies access and process published works.
The decision is seen as a major legal victory for tech firms across the country, setting a precedent for how courts may interpret copyright law in the age of large language models and generative AI.
However, the ruling wasn’t entirely in Anthropic’s favor. The court also found that the company stored over 7 million pirated books, a practice the judge said does violate copyright law. The infringement ruling opens the door to a trial set for December 2025, where the court will determine whether damages are owed to authors and publishers.
Two Truths in One Courtroom
The ruling strikes a complex balance between innovation and intellectual property protection. On one hand, the judge found that the transformative use of text data for AI training aligns with fair use doctrine, a stance that may influence similar lawsuits against other AI companies like OpenAI and Google.
On the other hand, storing and distributing unauthorized copies of full books—even if used for AI training—crosses the legal line, the court concluded.
Legal experts say the mixed ruling is likely to shape future AI policy and copyright reform debates, especially as AI systems increasingly rely on vast datasets scraped from the internet.
What Comes Next
The December trial will focus solely on damages. Plaintiffs, which include a group of authors and rights organizations, are expected to argue that Anthropic’s book hoarding has had a measurable financial impact on their work.
Meanwhile, the tech sector is viewing the fair use decision as a major shield in an increasingly litigious space. Whether Congress steps in to rewrite the rules of copyright for AI remains an open question.
