{"slug": "ai-firms-are-buying-and-pulping-rare-books-after-scanning-them-for-training-data", "title": "AI Firms Are Buying and Pulping Rare Books After Scanning Them for Training Data", "summary": "AI companies, including Anthropic, are buying and pulping rare out-of-print books to scan them for training data, a practice that U.S. District Judge William Alsup ruled legal in June 2025 under fair use. The ruling, part of the Bartz v. Anthropic case, allows the destructive scanning of purchased books, though Anthropic agreed to a $1.5 billion settlement over pirated digital copies. Booksellers in Europe and Australia report bulk orders, with suppliers sourcing up to a million books at a time, targeting pre-2022 titles to avoid AI-generated text.", "body_md": "*Antiquarian booksellers in Europe and Australia say AI companies, Anthropic among them, are quietly buying up out-of-print books by the thousand, slicing off their spines, scanning every page, and throwing away what's left. A court has already ruled the practice legal.*\n\nPieter de Vries runs a second-hand bookshop in Haarlem, in the Netherlands. In July, he got an order for 3,001 titles, mostly published between 2020 and 2021, from a buyer calling itself 2077AI and asking for the books to be shipped to China. \"I was shocked,\" he told Fortune, once he pieced together weeks later why anyone would want that exact list. It included academic titles from Emerald Publishing, Elsevier, Wiley, Routledge and Oxford University Press, the kind of backlist a publisher forgets it ever printed. Nobody wants three thousand engineering textbooks for the shelf. Somebody wants three thousand engineering textbooks for the dataset.\n\nAnthropic has been doing this for two years under an internal program called Project Panama. Court filings unsealed during the Bartz v. Anthropic copyright case describe the company hiring Tom Turvey, the former Google executive who ran Google Books, to buy physical books in bulk and feed them through a hydraulic cutting machine that shears off the binding so the loose pages can run through a high-speed scanner. One internal planning document, quoted in those filings, put the ambition plainly: \"Project Panama is our effort to destructively scan all the books in the world.\" Another read, less plainly but more honestly, \"we don't want it to be known that we are pursuing this project.\"\n\nIn June 2025, U.S. District Judge William Alsup ruled that this is legal. Digitizing a book you bought, then training a model on it, counts as fair use, he found in Bartz v. Anthropic, because the purchase itself was lawful and the resulting use is transformative. Anthropic still paid out separately, agreeing to a $1.5 billion settlement over pirated digital copies it had used before it started buying print. The buying-and-scanning part of the operation survived the lawsuit intact.\n\n## The trade elsewhere is watching\n\nDawn Albinger, president of the Australian and New Zealand Association of Antiquarian Booksellers and owner of Archive Books in Queensland, says no confirmed bulk AI order has landed in Australia yet. Her members are watching it happen to their American and European counterparts and bracing. \"The tyranny of distance and the cost of shipping is one reason why we haven't seen it yet,\" she said. It's a reprieve of geography, not of policy.\n\nLondon's rare book trade has had less luck staying out of it. One dealer told City AM they'd received two unusually large orders from anonymous email addresses with no company name attached, both buyers asking for extra photographs of the stock before committing to purchase. According to reporting from 404 Media, suppliers feeding AI labs have sourced anywhere from a thousand to a million books at a time, and buyers have specifically targeted books published before 2022, on the theory that anything older is \"structurally guaranteed\" to predate the flood of AI-generated text now polluting the web. That's the real driver here. The open internet is running out of clean human writing to train on, and a warehouse of pre-2022 print is one of the few remaining troves that's certifiably human.\n\n## What the ruling doesn't touch\n\nHere's the part that Alsup's ruling doesn't touch: once a book is cut apart and scanned, the physical object is gone. A first edition carries more than its text. Marginalia, an inscription, a particular binding, the paper stock a printer used in 1970, none of that survives a hydraulic spine cutter. Librarians and booksellers have been saying this since the court filings first surfaced, and it's not a sentimental objection. It's an archival one. A digitized page can be copied infinitely. The book it came from can't be un-pulped.\n\nThat's the tension nobody in the case actually resolved. Fair use answers a copyright question: did Anthropic have the right to train on this text. It doesn't answer a preservation question: should a rare or uncommon copy be destroyed to get that text, when a library or archive might have wanted the original later. Those are different questions. Only one of them made it in front of a judge. Booksellers like Albinger are left to raise the second one on their own, one anonymous bulk order at a time.\n\n**Also read:** [Google Built ChatGPT's Ancestor in 2018 and Watched Its Creator Leave Twice](https://startupfortune.com/google-built-chatgpts-ancestor-in-2018-and-watched-its-creator-leave-twice/) • [DeepSeek's V4-Flash Undercuts OpenAI and Anthropic on Price Again](https://startupfortune.com/deepseeks-v4-flash-undercuts-openai-and-anthropic-on-price-again/) • [Orchid's viral anniversary ad backfires into a debate over AI and intimacy](https://startupfortune.com/orchids-viral-anniversary-ad-backfires-into-a-debate-over-ai-and-intimacy/)", "url": "https://wpnews.pro/news/ai-firms-are-buying-and-pulping-rare-books-after-scanning-them-for-training-data", "canonical_source": "https://startupfortune.com/ai-firms-are-buying-and-pulping-rare-books-after-scanning-them-for-training-data/", "published_at": "2026-08-02 13:18:56+00:00", "updated_at": "2026-08-02 13:46:51.955883+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-policy", "ai-ethics", "ai-research"], "entities": ["Anthropic", "Pieter de Vries", "2077AI", "Emerald Publishing", "Elsevier", "Wiley", "Routledge", "Oxford University Press"], "alternates": {"html": "https://wpnews.pro/news/ai-firms-are-buying-and-pulping-rare-books-after-scanning-them-for-training-data", "markdown": "https://wpnews.pro/news/ai-firms-are-buying-and-pulping-rare-books-after-scanning-them-for-training-data.md", "text": "https://wpnews.pro/news/ai-firms-are-buying-and-pulping-rare-books-after-scanning-them-for-training-data.txt", "jsonld": "https://wpnews.pro/news/ai-firms-are-buying-and-pulping-rare-books-after-scanning-them-for-training-data.jsonld"}}