In his written ruling, Judge William Alsup explains that Anthropic did not begin by purchasing books. Instead, the company downloaded more than seven million copyrighted books from online piracy libraries, including Books3, LibGen and Pirate Library Mirror, to build what the court described as “a central library of all the books in the world” that Anthropic intended to keep indefinitely.
According to the ruling, the company later became, in the judge’s words, “not so gung ho” about using pirated books “for legal reasons.” Rather than deleting the unauthorized copies it had already obtained, Anthropic looked for another way to expand its collection using books it legally owned.
That effort led to an internal initiative known as Project Panama. To help develop it, Anthropic hired Tom Turvey, a former Google executive who had worked on the Google Books project. The court found that Turvey initially approached major publishers to explore licensing agreements for large collections of books. Those discussions, however, “withered,” and the company ultimately abandoned that approach.
Instead, Anthropic turned to book distributors and second-hand book suppliers, spending millions of dollars to purchase millions of printed books that it could digitize itself.
How the books were destroyed
The unsealed court exhibits describe an industrial process known as destructive scanning. Workers removed the bindings from newly purchased books so that individual pages could pass through high-speed scanners. Every page was converted into machine-readable text before the books were destroyed, leaving only digital copies in Anthropic’s internal library for AI training.
One internal planning document describes Project Panama as an effort to “destructively scan all the books in the world.” Another states, “We don’t want it to be known that we are working on this.” Those documents remained confidential until they were unsealed during the litigation.
Judge Alsup noted that the scanning process destroyed the original books, writing that “the source copy was destroyed.” The court found that Anthropic was not producing additional physical copies of the books but replacing a legally purchased printed copy with a digital version for computational use.
A legal distinction with major consequences
Project Panama became one of the central issues in the copyright lawsuit filed against Anthropic in the U.S. District Court for the Northern District of California.
In his June 2025 summary judgment ruling, Judge Alsup held that digitizing lawfully purchased books to train large language models constituted fair use under U.S. copyright law, even though the process destroyed the original books. The ruling concluded that converting purchased books into digital files for AI training was sufficiently transformative to qualify for fair use protection.
The judge reached the opposite conclusion regarding Anthropic’s earlier collection of pirated books. While the court accepted the digitization of books the company had legally purchased, it ruled that downloading and permanently retaining millions of unauthorized digital copies was not protected by the fair use doctrine.
Read also : US court approves record $1.5 billion settlement in landmark AI copyright case
The distinction means that the legality of AI training did not depend solely on how the books were used but also on how they were obtained.
The litigation later resulted in a $1.5 billion settlement covering claims related to Anthropic’s possession of pirated books. The settlement did not alter Judge Alsup’s earlier fair use ruling on Project Panama or his conclusion that scanning and digitizing lawfully purchased books for AI training was lawful under U.S. copyright law.
