Federal Court Finalized Anthropic Book Settlement
The ruling codifies a process for AI firms to scan purchased books before recycling the physical remains.
Updated on Oct. 4, 2026 in Artificial Intelligence
Live Poll
Should AI companies be permitted to destroy physical books to train their digital models?
In July 2026, a federal court finalized a class-action settlement involving AI developer Anthropic. The decision establishes legal parameters for AI companies to acquire and scan physical books for training data.
Why it matters
The settlement addresses how tech companies source training data from copyrighted physical works. It marks a formal legal reconciliation between AI model development and the use of physical texts for digital ingestion.
The settlement allows companies to use the first-sale doctrine to purchase books for digital scanning. The terms require firms to physically destroy book spines and recycle the remaining paper after the scanning process is complete.
The players
Anthropic
An AI research company focused on building large language models and developing techniques to align artificial intelligence systems with human intent.
Tom Congalton
A 74-year-old antiquarian bookseller who operates Between the Covers Rare Books, a business that has cataloged over 250,000 volumes.
The details
Companies secure bulk quantities of physical books to act as datasets for training large language models—algorithms capable of predicting and generating human-like text. Under the court-approved process, firms slice off the spines of books to facilitate rapid digital scanning by automated machines. Once digitized, the physical copies must be recycled, a process that removes the item from the secondary rare book market.
Timeline
1968: Tom Congalton began his career in book collecting.
2017: Author John Grisham acknowledged the Congaltons in his novel Camino Island.
July 2026: A federal court finalized the class-action settlement involving Anthropic.
The Tech Race
This settlement defines the boundary for how AI developers use physical assets to fuel model training. It sets a precedent that aligns data harvesting practices with the first-sale doctrine, effectively separating AI research infrastructure from traditional book preservation.
The ruling changes the workflow for AI developers who rely on mass book acquisition for model training by formalizing destruction requirements. Readers interested in rare book collections may see reduced physical inventory of older titles as industry demand for scanning sources persists.
The takeaway
The settlement formalizes the transition of printed media into digital training data while protecting the rights of companies to leverage physical goods. Watch for future court rulings that clarify whether the destruction of rare copies—some worth up to $250,000—triggers new intellectual property challenges.
Further reading
For more on how legal frameworks are shaping model development, visit the Artificial Intelligence section.
Source note: This article includes information reported by New Jersey Monthly.
Live Poll
Should AI companies be permitted to destroy physical books to train their digital models?
