A federal judge has handed down a mixed ruling that could reshape how artificial intelligence companies use copyrighted material to train their systems. The decision involves Anthropic, the company behind the Claude AI assistant, and marks the first time a court has directly addressed fair use in the context of generative AI.
U.S. District Judge William Alsup ruled Monday that Anthropic’s use of copyrighted books to train Claude was legal under fair use protections. However, the same judge found that the company violated copyright law by storing over 7 million pirated books in what he called a “central library.”
The case began when authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson filed a class-action lawsuit against Anthropic last year. They argued that the Amazon and Alphabet-backed company used pirated versions of their books without permission or payment to teach Claude how to respond to user questions.
What the Judge Got Right About AI Training
Judge Alsup sided with Anthropic on the core question of whether using copyrighted works to train AI systems constitutes fair use. He described the training process as “exceedingly transformative,” comparing it to how human writers learn from reading other works.
“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 in his decision.
This reasoning aligns with Anthropic’s defense that their AI system copied books to “study Plaintiffs’ writing, extract uncopyrightable information from it, and use what it learned to create revolutionary technology.” The company argued that copyright law actually encourages this type of learning because it promotes human creativity.
Where Anthropic Crossed the Line
While the judge approved of the training method, he drew a sharp line at how Anthropic obtained and stored the books. The company had created what Alsup called a “central library of all the books in the world,” containing millions of pirated copies that weren’t necessarily being used for AI training purposes.
This distinction proved crucial. Alsup questioned why any company would need to download books from pirate sites when they could have “purchased or otherwise accessed lawfully” the same material. He rejected Anthropic’s argument that the source of the books was irrelevant to fair use analysis.
The decision will see Anthropic go on trial in December to decide on copyright damages. Wilful copyright infringement in U.S. law involves statutory damages of up to $150,000 per work, leaving the company with huge financial exposure.
Wider Implications for the AI Sector
The ruling arrives as a number of AI companies are being sued in the same fashion by authors, news outlets, and other copyright holders. OpenAI, Microsoft, and Meta Platforms are all battling claims of improper use of copyrighted material to train their AI models.
The divided judgment has something for both sides to rejoice and dread. AI companies can boast about the judge’s endorsement of transformative training practices, and copyright holders can highlight the stringent limits on how companies obtain their training material.
For Anthropic, the firm was pleased with the major decision. A spokesperson stated that they were thrilled that the court acknowledged their AI training as “transformative” and “consistent with the aim of copyright to foster creativity and encourage scientific progress.”
What’s to Come for Anthropic
The trial in December will only decide damages for storing the pirated book, not the more general issue of training practices for AI. But the impact of this ruling will extend to other cases in the industry pending.
The ruling also emphasizes the need for AI businesses to responsibly source the training data. Although the ruling indicates learning from copyrighted work may be considered fair use, it leaves one thing absolutely clear: companies cannot simply borrow material from wherever it is easiest to obtain it.
As the industry for AI expands, the case teaches us that courts will be eager to make distinctions between legitimate transformative uses and pure copying that is more than is necessary for innovation. Businesses will have to be more cautious about what they copy and how they acquire it.
