Judge Rules Anthropic’s Use of Purchased Books to Train AI Is Fair Use

A federal judge has ruled that Anthropic did not violate copyright law by using legally purchased books to train its artificial intelligence models, marking a significant, if narrow, legal victory for the AI startup. However, the company still faces a separate trial over allegations that it relied on pirated books in the same training process.

The decision, issued by Judge William Alsup of the Northern District of California, concluded that scanning and digitizing physical books bought by Anthropic — even after physically disassembling them — qualifies as fair use. The court also found that using those scanned books to train large language models (LLMs) was “sufficiently transformative” to fall under the same protection.

This is the first time a U.S. court has clearly backed the AI industry’s argument that model training on legally obtained material can be fair use. Still, the ruling is limited to a specific scenario: books Anthropic physically purchased and digitized for training purposes.

Legal Claims Split in Two

The ruling stems from a lawsuit filed by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, who accused Anthropic of training its Claude family of AI models on pirated versions of their books. While Judge Alsup sided with Anthropic on the purchased books, he declined to dismiss claims related to pirated content.

According to the court, Anthropic built a centralized digital library containing both purchased and pirated books. The company’s defense — that some of these materials may not have been directly used for training — didn’t convince the judge that such copying could be fair use.

“This order doubts that any accused infringer could ever meet its burden of explaining why downloading source copies from pirate sites that it could have purchased or otherwise accessed lawfully was itself reasonably necessary to any subsequent fair use,” Alsup wrote, emphasizing that point in his ruling.

Authors’ Rights vs. AI’s Progress

Judge Alsup drew a sharp line between protecting authors from unfair exploitation and enabling transformative innovation. In his opinion, he compared Anthropic’s use of books for AI training to educators teaching children to write — a process that inherently involves learning from existing work.

“Authors’ complaint is no different than it would be if they complained that training schoolchildren to write well would result in an explosion of competing works,” he wrote, adding that the Copyright Act is meant “to advance original works of authorship, not to protect authors against competition.”

Anthropic welcomed the ruling, emphasizing its models were designed to create novel outputs, not to replicate or replace the original books.

“We are pleased that the Court recognized that using works to train LLMs was transformative — spectacularly so,” said Anthropic spokesperson Jennifer Martinez. She added that the company’s use of content was consistent with copyright law’s aim to “enable creativity and foster scientific progress.”

Looking Ahead

While the fair use ruling offers some legal clarity for AI developers relying on purchased materials, the unresolved claims around pirated books pose a serious challenge. A separate trial will determine whether Anthropic’s use of unauthorized copies was unlawful and what damages, if any, the company may owe.

This case may shape future legal standards for how AI models can be trained — and where the line lies between innovation and infringement.

Weighing the Trade-Offs

Even as courts begin to define the legal boundaries for AI training, the rulings may come too late to meaningfully rebalance the landscape. Companies like Anthropic already trained models on massive datasets, including potentially pirated materials. A fair use decision on legally purchased books sets precedent, but it doesn’t reverse any benefits gained from earlier, more controversial practices.

If courts ultimately impose fines or damages for unauthorized use, it raises a broader question: are those penalties enough to deter future violations — or simply the cost of doing business? The line between fair innovation and unfair advantage may be getting clearer, but for many authors and rights holders, the harm may already be done.

But there’s also a broader view. To achieve the remarkable capabilities we now expect from AI — from instant translation to life-saving medical insights — this messy phase may have been unavoidable. The outputs rely on enormous volumes of data, much of it human-made. While rights holders deserve compensation, the long-term public benefits of powerful AI systems may, in some eyes, justify how we got here.

The question isn’t just what’s legal — it’s what kind of trade-offs we’re willing to live with.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.