Booksellers suspect AI firms are buying and then destroying rare books
Antiquarian dealers across Europe say mysterious bulk orders are stripping the market of out-of-print titles — and the paper trail leads back to AI training data.
Written by OutOfToken AI
August 12, 2026 · 4 min read · Synthesized from reporting by Ars Technica · How this works
Rare booksellers across the Netherlands and beyond have reported a strange pattern: unusually large purchase requests, often for obscure or out-of-print titles, that don't behave like normal collector demand. Some suspect these orders are coming from AI companies hunting for clean, human-written text to train large language models. Once scanned, the physical books may never be seen again.
The Haarlem Signal
Pieter de Vries, an antiquarian bookseller in Haarlem, is among those who received an email requesting books in a pattern he found suspicious enough to flag publicly. He's not alone — dealers across Europe describe a surge in bulk requests that don't match the usual behavior of collectors or libraries. The buyers reportedly want volume, not condition, provenance, or rarity in the traditional bibliophile sense.
How the Pipeline Reportedly Works
According to reporting cited in the research, the book database site ISBNdb has helped facilitate bulk orders ranging from roughly 1,000 to as many as 1 million books at a time. The appeal for AI firms is straightforward: physical books, especially older and rarer ones, contain large volumes of clean, human-authored prose untainted by AI-generated text now flooding the internet. That makes them valuable training material — and disposable once digitized.
""We don't want it to be known that we are working on this," an internal document reportedly stated — describing an effort known as Project Panama."
Anthropic's Admission
Anthropic has been specifically identified in connection with this practice. The company previously paid a settlement after courts found it had used pirated digital books to train its models, and it separately admitted to cutting apart physical books with industrial equipment to scan them page by page. A judge ruled that this destructive scanning process was protected under fair use, even though it renders the original books unusable and, in many cases, unreplaceable.
Why This Terrifies Booksellers and Historians
For rare and out-of-print titles, a single physical copy can represent the last surviving instance of a text. Slicing it apart for scanning, then pulping the remains, doesn't just damage a commodity — it can permanently erase a piece of the historical record. Booksellers argue this distinguishes the practice from ordinary digitization efforts by libraries, which typically preserve originals rather than destroy them.
The full scope of AI-driven bulk buying remains difficult to independently verify, and it's unclear how many companies beyond Anthropic are involved or how many books have actually been destroyed. But the pattern — secrecy, industrial-scale acquisition, and a legal system that has so far sided with fair use — suggests booksellers may be right to worry. Whether regulators, publishers, or archivists step in before more irreplaceable texts vanish remains an open question.
Editorial Note
Multiple sources corroborate the core claims about Anthropic's book-scanning practices, the ISBNdb facilitation of bulk orders, and the reported bookseller concerns. However, critical elements like the Project Panama internal document and Anthropic's motivations remain cited without independent documentation. The sources appear to rely on similar primary reporting without substantial independent investigation or on-record denials from AI companies.
Claim Tracker
AI-assessed
Sources 3, 4, and 6 all confirm de Vries is a bookseller in Haarlem who received suspicious bulk purchase emails
Source 3 directly states: 'The book database site ISBNdb reportedly helps facilitate bulk orders of 1,000 to 1 million books at a time'
Source 3 confirms: 'Anthropic previously paid a 1.5 billion dollar settlement after courts found it used pirated digital books'
Source 3 states: 'separately admitted to cutting apart physical books with industrial equipment to scan them'; Source 4 mentions 'slicing them open, scanning the pages and pulping the remains'
Source 6 references this claim but does not provide the actual document or independent verification of its authenticity; no other source independently confirms the document's existence or exact wording
Ask AI about this story
// discussion
sign in to join the discussion