Full Breakdown
AI Companies Accumulate Bulk Secondhand Books for Training Data, Booksellers Report
8/15/2026, 8:09:03 PM
Unusual Bulk Purchases Disrupt Secondhand Trade
Secondhand booksellers in the UK, Ireland, the US, Canada, Australia and Europe have seen a surge of large, thematically incoherent orders paid at “top whack” prices and shipped to a postcode near Heathrow Airport. The pattern began three months ago and continues through the summer, prompting speculation that AI firms are buying the titles to create “clean” training corpora.
Background & Context
AI developers need vast amounts of text free of AI-generated content. Books published before 2022 are prized because they have not been altered by chat-bots. Historically, large-language-model training has incorporated scanned books, and the current wave may represent a renewed effort to harvest older titles.
Key Figures & Groups
- Stuart Manley – co-owner, Barter Books, Alnwick.
- Jim Shaughnessy – owner, MW Books, Claregalway.
- David Gower-Spence – owner, BookLovers of Bath.
- Anthropic – AI company behind Claude.
- Zoom Books – Canada-based “book recycling” firm.
Timeline
- January – A UK bookseller notes orders for roughly 6,000 books shipped to the Heathrow-area postcode.
- May – MW Books and others begin receiving large, varied orders.
- May-July – BookLovers of Bath records about 200 titles.
- Three months ago – Barter Books observes the first wave of “strange combination” orders.
- Recent months – Australian booksellers report comparable spikes.
Data & Statistics
- One seller sold “hundreds” of random books to three buyers for about £4,000 (? $5,300).
- Another reported a bulk order equal to a typical seven-day sales volume of 2–3 thousand books.
- A seller handling orders since January has purchased roughly 6,000 titles, amounting to “thousands of pounds.”
Impact on Book Trade
Booksellers say the influx is “disrupting the secondhand bookselling business” because the purchases ignore normal thematic patterns and bypass discount negotiations. The concentration of deliveries at a single logistics hub raises concerns about the future availability of rare and out-of-print volumes.
Official Statements & Responses
Anthropic responded that “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.”
On-the-Ground Reports
Manley described an order that included an Estonian translation of John le Carré’s *The Mission Song*, an imprint of Anne Brontë’s *Agnes Grey* and the October 1983 issue of *Warship* magazine. Shaughnessy highlighted subjects ranging from 18th-century African agricultural implements to 1950s racing-car driver biographies. Gower-Spence said the bulk purchases were “overwhelming for a while” but that he remained “relatively well organised.”
Conflicting Reports & Gaps
Anthropic’s statement asserts its programs do not purchase or destroy rare books. No independent verification of the “Project Panama” name or its scope has been provided, and the identities of the buyers—whether directly affiliated with Anthropic or third-party recyclers—remain unconfirmed.
Verbatim Quotes
- “It’s definitely a strange combination of books,” — Stuart Manley, Barter Books
- “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
- “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, BookLovers of Bath
