AI Companies Bulk Buy and Destroy Used Books to Train Models
AI companies reportedly purchased tens of thousands of used books globally through intermediaries including ZoomBooks and PrepFort, scanned them for clean training data, and then destroyed them. The practice has depleted used bookstore inventories and targeted pre-2022 publications specifically to obtain text free from AI-generated content.
What happened
Booksellers worldwide have reported coordinated bulk purchases of used books by intermediary companies that appear to be automated. A Canadian company called ZoomBooks has purchased tens of thousands of used books from Spanish, German, American, New Zealand and Australian booksellers, with orders of hundreds or thousands of copies with no regard to price or thematic coherence.
The books are sent to PrepFort, a logistics operator in Illinois, where they are scanned and catalogued. ZoomBooks denies directly collaborating with Anthropic but describes itself as a used bookstore focused on recycling. The books are subsequently destroyed after being scanned for training data.
Reported impact
The practice has depleted inventories at used bookstores globally. A US bookseller reported a shift from selling twenty books in a good week to hundreds following the arrival of bulk buyers. Books targeted include out-of-print non-fiction works, technical manuals, historical records and publications from before 2022 that are free from AI-generated content.
Index analysis
Used books published before the widespread deployment of generative AI contain training data free from AI-generated text, known as 'AI slop'. Publishers and booksellers report that AI companies have specifically targeted such publications for purchase and destruction after scanning, citing fair use compliance.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.