A federal judge ruled that AI company Anthropic did not break copyright law when it used books to train its Claude large language model (LLM).

In a summary judgment order released on Monday, U.S. District Judge William Alsup said the company’s use of copyrighted works by Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson was protected under the “fair use” doctrine. 

In copyright law, fair use often depends on whether the new use is “transformative,” meaning it creates something new rather than replacing the original. 

“The purpose and character of using copyrighted works to train LLMs to generate new text was quintessentially transformative,” Alsup wrote. “Like any reader aspiring to be a writer.”

“The technology at issue was among the most transformative many of us will see in our lifetimes,” Alsup further added.

Part of the lawsuit focuses on a cache of approximately 7 million pirated books that Anthropic retained in a “central library.” Although the company ultimately chose not to use these materials to train its language models, the court raised concerns about their storage and acquisition.

Alsup has ordered a trial to determine how these pirated books were used in creating the central library and to assess any resulting damages.

The recent decision is a major moment for Anthropic, which faces several lawsuits over how it collects and uses copyrighted material to train AI systems. In June, social media platform Reddit filed a lawsuit against Anthropic, accusing it of illegally scraping user comments from the platform to train Claude.

A spokesperson for Anthropic told Wired that the company was “pleased” with the recent court ruling and that it supports copyright’s purpose of “enabling creativity and fostering scientific progress.”