TL;DR

Second-hand and antiquarian booksellers across Europe are being approached to supply books in bulk to intermediaries acting for AI companies. The volumes run to thousands of titles per enquiry. The books are fed through high-speed scanners that cut off the spines, and the originals are then pulped. The practice was exposed this week by 404 Media, which reported that one service, ISBNdb, offered orders of up to a million books at a time.

Pieter de Vries, an antiquarian dealer in Haarlem who typically sells one book a day, was sent a spreadsheet of some 3,000 titles in mid-June and took it for a fraud. Marçal Font i Espí, a bookseller and university literature teacher near Barcelona, noticed unusual bursts of orders where postage exceeded the books’ value, and reports the same pattern from colleagues in Spain, France, Germany and New Zealand. ISBNdb told the Telegraph the bulk service was a test of demand it never launched, and acknowledged in its own blog that the optics were a problem.

The legal mechanics are the story. A California ruling last month in the authors’ case against Anthropic held that scanning a book and destroying the physical copy is transformation rather than duplication, because the digital version replaces the original. Destruction therefore strengthens the legal position. Buying second-hand stock and shredding it is now cheaper and safer than negotiating publisher licences — an incentive created by copyright law rather than despite it.

Anthropic hired a former Google Books executive in February 2024 with an internal remit to obtain the world’s books with minimal legal and commercial friction, and spent millions buying and destroying stock. It declined to comment.

Ed Newton-Rex of Fairly Trained argues a $1 second-hand purchase should not confer the right to train a commercial model that then competes with the author. Cornell law professor James Grimmelmann notes copyright treats a mass-market paperback and one of five surviving copies identically. Even Elon Musk has said SpaceX should scan rare books without cutting them.

For UK publishers, libraries and authors this lands directly on the live copyright question. Britain’s consultation on text and data mining exceptions has run for over a year without resolution, and the US ruling shows what a permissive settlement produces in practice.

Looking Forward

The immediate risk is irreversible: where a bought copy is the last accessible one, the work survives only as a scan inside a private training corpus. Whether the UK writes a licensing route into law, or follows the American position by default, determines which of those outcomes becomes standard here.