Judge OKs AI Training on Lawful Books; $1.5B Settlement for Pirated Copies
Judge Alsup ruled AI training on legally purchased books is fair use; pirated books led to $1.5B settlement.
Why it matters: The ruling clarifies how AI developers can use copyrighted works legally, impacting authors and publishers. Meanwhile, concerns rise over AI firms destroying physical books to create training data.
- June 2025: Judge William Alsup ruled using lawfully acquired copyrighted books for AI training is fair use.
- August 2026: Anthropic settled for $1.5 billion over use of pirated books from unauthorized online libraries.
- AI companies reportedly buy and shred millions of books to generate training data, prompting FTC investigations.
- More than a dozen groups urged the FTC in August 2026 to examine if book destruction is an unfair market practice.
In a key 2025 ruling, U.S. District Judge William Alsup decided that Anthropic’s use of copyrighted books legally purchased for training its AI model Claude qualifies as fair use under copyright law. He described this use as “transformative,” since it repurposed the material for AI development rather than traditional reading (Los Angeles Times, TechSpot).
However, the judge distinguished this from Anthropic’s downloading and retaining copyrighted works from unauthorized online libraries like Library Genesis, often called "shadow libraries," where books are shared without permission. Such use was deemed “irreparably infringing,” leading to Anthropic’s $1.5 billion settlement with authors in August 2026 to resolve copyright claims (Creative Bloq).
At the same time, several AI firms have been reported to purchase physical books in large quantities and destroy them, typically by shredding, to extract content for training datasets. This controversial practice raises legal and ethical concerns about market harm and fair competition since the physical books become unusable. In August 2026, more than a dozen civil society organizations petitioned the Federal Trade Commission (FTC) to investigate these destructive methods as potentially unfair business practices (Tom's Hardware, Axios).
Judge Vince Chhabria, who has overseen related AI copyright cases, clarified that rulings like Alsup’s do not broadly legalize all AI training on copyrighted works. He noted that many plaintiffs find it hard to show "market dilution" — a legal claim that the copyrighted market is harmed — but warned that stronger evidence could change case outcomes, especially against large AI companies such as Meta (TechSpot).
These developments highlight the complex balance courts and regulators must strike between enabling AI innovation and protecting creators' rights as AI training data sourcing faces growing scrutiny.
By the numbers:
- $1.5 billion — settlement amount Anthropic agreed to over pirated book use
- June 2025 — date of Judge Alsup’s fair use ruling for lawful AI training materials
- August 2026 — month when FTC received petitions concerning AI firms’ book destruction
Yes, but: While the ruling favors fair use for legally obtained books, the legality and ethics of buying and destroying physical books to train AI remain unsettled, inviting regulatory scrutiny and potential future litigation.
What's next: FTC is expected to investigate the book destruction practices of AI companies, with outcomes that could shape future regulatory measures.