AI Companies Buy Bulk Secondhand Books for Model Training

Secondhand booksellers across the United Kingdom, Ireland, the United States, Australia, and Europe report a surge in unusual bulk orders for non-thematic books. Sellers describe the orders as sporadic and potentially machine-driven, with buyers paying premium prices for diverse titles ranging from 18th-century agricultural texts to 1950s racing biographies.

Booksellers suspect AI companies are acquiring these physical copies to scan contents for training large language models, specifically targeting pre-2022 texts to avoid data contaminated by chatbot-generated content. These orders are typically placed through marketplaces such as Biblio and AbeBooks. Individual stores, including Barter Books in the UK and MW Books in Ireland, report selling hundreds of random titles to mystery buyers.

Anthropic PBC, the developer of the Claude chatbot, has acknowledged that sourcing books is a common industry practice for training models. The company has reportedly spent tens of millions of dollars acquiring books for scanning and recycling. While Anthropic PBC confirms it uses a mix of web data and commercially acquired datasets, it denies destroying rare or antiquarian books during its acquisition programs.