Secondhand book sales are booming. Is it because of AI?
Independent second-hand booksellers are reporting unusually large bulk orders, prompting suspicion that the books are being bought to train artificial intelligence systems rather than for readers. The trend matters because some books may be digitised through “destructive scanning”, where their spines are removed and pages scanned before the remains are recycled.
The speculation follows a 2025 US ruling that Anthropic’s use of legally purchased books for AI training did not breach copyright, with court documents later revealing its “Project Panama” aimed to scan books at industrial scale. Barter Books in Northumberland received one Canadian bulk order equivalent to a week’s normal sales, while other sellers have seen similarly unexplained purchases of varied titles, from Latin texts to cowboy novels. Experts say such diversity could be valuable for AI training, although UK copyright law generally requires permission from rights holders for copying and training.
In short
- Unusual bulk purchases may be supplying books for AI training.
- Some purchased books could be dismantled and scanned.
- UK copyright rules are stricter than the US approach.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
The surge in bulk purchases could plausibly be linked to AI companies or intermediaries seeking large, diverse corpora of printed material that are not readily available in licensed digital form. Supporters of this explanation note that unusual order sizes and titles may fit the needs of data collection at scale, while the economic incentives to digitise cheaply can be substantial. They would argue that understanding this supply chain matters because authors, publishers and booksellers deserve transparency about whether their work is being reproduced for commercial AI training and then discarded.
The case against
Bulk secondhand-book orders do not by themselves establish an AI connection, as libraries, resellers, collectors, film and television productions, educational projects and other commercial buyers can also create abrupt demand. Even if books are being digitised, it may be premature to assume they are being pulped or used unlawfully without evidence about the purchasers, scanning process and rights arrangements. A cautious view holds that attributing a complex market trend to AI risks creating an appealing but unverified narrative, and that the claim should be tested through documented reporting rather than inference from unusual sales patterns.
Explained simply
Artificial intelligence systems learn patterns from large collections of text, including books. To use printed books as training material, a company may scan their pages; in destructive scanning, the spine is cut off and the pages are scanned before the book is recycled.
In 2025, a US court ruled that Anthropic’s use of lawfully bought books to train AI did not breach copyright in that case. The ruling followed revelations about Project Panama, a plan to scan books on a large scale. UK copyright rules generally require permission from rights holders to copy books for AI training, so the legal position differs between the two countries.
AI Americas Entertainment Technology World