AI Companies’ Bulk Book Purchases Spark Concern Among Secondhand Sellers

Anthropic, the developer of the Claude language model, has been buying secondhand books in bulk to augment its training data. “Claude is trained on a mix of publicly available web data, commercially acquired datasets, and data we generate ourselves. Sourcing books is a widely used approach for training large language models across the AI industry. None of our data acquisition programs buy and destroy rare or antiquarian books.” — Anthropic

The Guardian

Printed works that are not digitized online provide fresh language data that can improve model performance. Recent EU regulations, coming into force this month, have increased pressure on AI developers to document data provenance, prompting firms to secure clearly licensed sources such as physical books.

The Guardian

• Two dozen booksellers in the UK and Ireland reported bulk orders beginning three months ago.

The Guardian

• Barter Books in Alnwick sold “hundreds” of random titles to three buyers for roughly £4,000. “It’s definitely a strange combination of books,” — Stuart Manley.

The Guardian

• MW Books in Claregalway noted “sporadic, presumably machine-driven” orders covering topics from 18th-century African agriculture to 1950s racing-car biographies. “Personally, I don’t see we have the right to dictate what any customer does with their purchases but it is more than curious to see what’s being selected.” — David Gower-Spence.

The Guardian

• BookLovers of Bath received about 200 books between May and July, many priced at “top whack” with no discount. “It was overwhelming for a while and I’m relatively well organised,” — Gower-Spence.

The Guardian

• One seller traced a postcode to freight warehouses near Heathrow that had received 6,000 books since January, paying premium rates despite the lack of thematic coherence. “My theory is that it’s masses and masses of AI money being squandered.” — Stuart Manley.

The Guardian

Anthropic reiterated that its data-acquisition practices do not target rare or antiquarian volumes and that book sourcing is a standard industry method. The company repeated its earlier statement about the mix of data sources used for Claude.

The Guardian

Sellers describe the orders as “strange” and lacking logical theme, suggesting they are driven by AI firms’ need for raw text rather than market demand. The lack of transparency about the ultimate buyers has led some to label the activity “overwhelming.”

The Guardian

No party has confirmed the exact volume of books purchased for AI training, nor identified most of the titles beyond a few examples (e.g., an Estonian translation of *The Mission Song* and the 1983 edition of *Warship*). Anthropic’s statement does not disclose quantities or pricing, leaving the scale of the operation unclear.

The Guardian