# Anthropic settles book piracy case for $1.5B, exposing secret Project Panama > Source: > Published: 2026-07-27 23:55:42+00:00 A federal judge on July 20th approved a $1.5 billion copyright settlement covering pirated books acquired by [Dario Amodei (@DarioAmodei)](https://x.com/DarioAmodei)-led [Anthropic](https://www.anthropic.com/), closing the case that exposed the AI developer's secretive Project Panama book-scanning operation. The final order arrived years after Anthropic's founders concluded that books were among the most cost-effective sources for building a competitive large language model. Anthropic, founded in 2021 and led by siblings Dario and Daniela Amodei, originally presented its mission as developing AI systems that were "steerable, interpretable, and robust," according to [Anthropic's Series A announcement](https://www.anthropic.com/news/anthropic-raises-124-million-to-build-more-reliable-general-ai-systems). Court records show how that mission collided with the data requirements of frontier-model development. Anthropic first assembled a large digital library from Books3 and pirate repositories including Library Genesis and Pirate Library Mirror. Anthropic later shifted toward buying physical books, removing their bindings, scanning their pages and discarding the paper copies. A [July 27th post on X](https://x.com/itsolelehmann/status/2081838425276383662?s=46) from [Ole Lehmann (@itsolelehmann)](https://x.com/itsolelehmann) recirculated language from an internal planning document that described the physical acquisition effort as Project Panama. The document said Anthropic wanted to "destructively scan all the books in the world" and did not want the effort publicly known, according to [The Washington Post's January report](https://www.washingtonpost.com/technology/2026/01/27/anthropic-ai-scan-destroy-books/). ### From pirate libraries to purchased books The acquisition campaign began early in Anthropic's history. A [June 23rd, 2025 court order](https://www.washingtonpost.com/documents/cf8f6674-7b14-4b7d-ab6f-bbd9ad45963f.pdf) found that co-founder [Ben Mann (@8enmann)](https://x.com/8enmann) downloaded at least 5 million books from LibGen in June 2021. Anthropic downloaded at least 2 million more from Pirate Library Mirror in July 2022, on top of 196,640 Books3 titles acquired earlier. Judge William Alsup concluded that Anthropic had pirated more than 7 million book copies. The order said Anthropic retained the files as a general-purpose research library, including books Anthropic had decided against using to train its models. Anthropic changed tactics as its lawyers grew concerned about training on pirated material. In February 2024, Anthropic hired Tom Turvey, who had led partnerships for Google's book-scanning project, and assigned him to obtain a comprehensive book collection while limiting the accompanying legal and commercial work. Turvey briefly contacted publishers about licensing books for AI training, according to the order, before Anthropic pursued bulk purchases from distributors and retailers. Anthropic spent millions of dollars buying millions of print books, many of them used. Contractors removed the bindings, cut the pages, produced searchable PDF files and discarded the physical copies. That distinction determined the legal outcome. Alsup held that training language models on copyrighted books was a transformative fair use. He also found that converting a lawfully purchased print book into an internal digital copy was fair use when the original was destroyed and the digital version was not distributed. The earlier pirate downloads received no such protection. Alsup found that building and retaining a central library from unpaid copies was a separate, non-transformative use. The prospect of a trial over those files and potential statutory damages pushed Anthropic toward settlement. ### The settlement draws a line around acquisition The settlement covers 482,460 books obtained from LibGen and Pirate Library Mirror. As of April 16th, authors and publishers had claimed interests in 440,490 works, or 91.3% of the list, according to the [final approval order](https://lunch.publishersmarketplace.com/wp-content/uploads/2026/07/SettlementApproval.pdf). The estimated payment is roughly $3,000 per book before final allocation adjustments. Anthropic must also destroy original files downloaded from LibGen and Pirate Library Mirror, along with copies derived from those repositories, subject to evidence-preservation requirements. Anthropic has maintained that the pirate datasets were not used in any commercially released model's training corpus. Anthropic deputy general counsel Aparna Sridhar told the [Associated Press](https://apnews.com/article/ai-anthropic-copyright-settlement-claude-books-bartz-74b140444023898aeba8579b6e9f0d63) that the June 2025 ruling established AI training on books as fair use and said Anthropic was preparing to close the matter. The release is narrower than a blanket settlement of Anthropic's relationship with authors. It covers claims arising from past acquisition and copying through August 25th, 2025. It does not release claims based on model outputs or future conduct. Project Panama therefore survived legal scrutiny as a method of creating a private digital library from purchased books. Anthropic's unpaid digital acquisitions generated the $1.5 billion liability. Together, the two tracks show the price frontier AI developers placed on books and the legal difference between buying training material and copying it from a pirate repository.