Anna’s Archive Rallies Book Scanners, Accuses AI Firms of Hoarding Data
Demand for high-quality text has pushed the generative-AI data supply chain beyond web scraping and into physical books. Anna’s Archive, the world’s largest shadow library, says Anthropic and other developers have bought books in bulk, removed their bindings, scanned the pages and destroyed the originals to build proprietary, clean training corpora. The dispute also reflects a broader copyright battle over model developers’ past use of collections obtained from pirate libraries including LibGen and PiLiMi.
Anna’s Archive has called on volunteers to scan and upload books from their own collections, framing the campaign as an effort to preserve material consumed by destructive digitization. The legal picture is more nuanced: U.S. District Judge William Alsup ruled on June 23, 2025, that Anthropic’s scanning and destruction of lawfully purchased books for a digital research library was transformative fair use. The ruling did not broadly absolve the company for retaining millions of books acquired from pirate sources.
All Coverage
2 original reportsThe Backstory
The history behind this eventAnthropic's Secret Project Panama Book-Scanning Program Revealed After $1.5 Billion Copyright Settlement
Anthropic launched a physical-book digitization program code-named Project Panama to obtain high-quality training data. It bought books, removed their bindings, scanned them and destroyed the paper copies for use in training its Claude models. The case highlights the legal boundaries surrounding AI training data: repurposing lawfully purchased books may qualify as fair use, while obtaining files from pirated sources may constitute infringement.
Unsealed documents disclosed on January 27, 2026, showed that contractors had planned to process 500,000 to 2 million books within six months. On September 5, 2025, Anthropic agreed to establish a settlement fund of at least $1.5 billion covering about 465,000 works, or an estimated $3,000 per work. The court granted preliminary approval on September 25 and completed its final fairness hearing on May 14, 2026, but a ruling on final approval remains pending.
Subscribe to Mark Radar Weekly
Every Friday, the week's strongest signals in your inbox. Unsubscribe anytime.
If you search news on Google, you can set Mark Radar as a preferred source—our coverage will show up more often in your results. Set as preferred source on Google →