{"slug": "what-the-evidence-actually-shows-about-ai-companies-destroying-books", "title": "What the Evidence Actually Shows About AI Companies Destroying Books", "summary": "Court records establish that Anthropic spent millions of dollars acquiring millions of print books, then had vendors remove their bindings, scan their pages and discard the originals, according to a district judge's order. A separate investigation by 404 Media identified ISBNdb as a broker marketing bulk physical-book acquisition to AI developers, but did not identify customers or confirm completed transactions. The strongest evidence of destruction remains Anthropic's own court record, which also shows a judge found the one-for-one conversion of purchased print copies into internal digital copies to be fair use.", "body_md": "# What the Evidence Actually Shows About AI Companies Destroying Books\n\n- Court records establish that Anthropic spent millions of dollars acquiring millions of print books, then had vendors remove their bindings, scan their pages and discard the originals.\n[[1]](https://www.fbm.com/content/uploads/2025/07/6-23-25-Alsup-Anthropic-Order.pdf) - ISBNdb markets bulk physical-book acquisition to AI developers, with reported order sizes between 1,000 and 1 million books, but its customers and completed transaction volumes remain undisclosed.\n[[2]](https://www.404media.co/ai-companies-are-buying-tons-of-old-books-because-theyre-free-of-ai-slop/) - Booksellers have reported unusual orders for obscure titles, but those accounts do not prove the buyers were AI labs or that the books were destroyed.\n[[2]](https://www.404media.co/ai-companies-are-buying-tons-of-old-books-because-theyre-free-of-ai-slop/)[[3]](https://charliebecker.substack.com/p/is-an-ai-company-buying-up-all-the) - A district judge found Anthropic’s one-for-one conversion of purchased print copies into internal digital copies to be fair use. A separate $1.5 billion settlement covered pirated digital books, not the purchased-and-scanned copies.\n[[1]](https://www.fbm.com/content/uploads/2025/07/6-23-25-Alsup-Anthropic-Order.pdf)[[4]](https://law.justia.com/cases/federal/district-courts/california/candce/4%3A2024cv05417/434709/680/)\n\nReports that artificial-intelligence companies are secretly buying and shredding millions of books combine one firmly documented operation with a newer market whose participants remain mostly hidden.\n\nThe documented operation belongs to Anthropic. Court records show that the Claude developer bought millions of physical books, paid service providers to strip their bindings and cut their pages, scanned them into searchable files and discarded the paper originals. Anthropic spent millions of dollars building the resulting private research library.[[1]](https://www.fbm.com/content/uploads/2025/07/6-23-25-Alsup-Anthropic-Order.pdf)\n\nThe broader claim comes from a July 21 investigation by 404 Media titled “AI Companies Are Buying Tons of Old Books Because They’re Free of AI Slop.” It identified ISBNdb as a broker marketing high-volume physical-book acquisition to AI developers. According to the investigation, ISBNdb offers to source between 1,000 and 1 million books per engagement and promotes older print as human-written material uncontaminated by generative-AI output.[[2]](https://www.404media.co/ai-companies-are-buying-tons-of-old-books-because-theyre-free-of-ai-slop/)\n\nThat reporting documents a business designed to supply print books to AI companies. It does not identify the customers, confirm that million-book orders were completed or establish what happened to books acquired through the service. The strongest evidence of destruction remains Anthropic’s own court record.\n\n## Project Panama left a detailed paper trail\n\nAnthropic hired Tom Turvey, the former head of partnerships for Google’s book-scanning project, in February 2024. The court’s summary-judgment order says he was tasked with obtaining “all the books in the world.” After limited contacts with publishers about licensing, his team approached distributors and retailers about bulk purchases for an Anthropic “research library.”[[1]](https://www.fbm.com/content/uploads/2025/07/6-23-25-Alsup-Anthropic-Order.pdf)\n\nAnthropic ultimately bought millions of books, often used and sometimes in batches of tens of thousands, from sellers including Better World Books and Britain’s World of Books. The final number and total cost remain redacted. A proposal from a vendor that worked on the project said Anthropic wanted capacity to convert between 500,000 and 2 million books over six months. The Washington Post reported, after reviewing more than 4,000 pages of unsealed filings, that the company spent tens of millions of dollars during the broader operation.[[5]](https://www.washingtonpost.com/technology/2026/01/27/anthropic-ai-scan-destroy-books/)\n\nThe proposal described using a hydraulic cutting machine to remove the bindings, production-grade scanners to digitize the pages and a recycling company to collect the remains. The district court separately found that Anthropic’s service providers stripped the books from their bindings, cut the pages to size and discarded the paper originals after scanning.[[1]](https://www.fbm.com/content/uploads/2025/07/6-23-25-Alsup-Anthropic-Order.pdf)[[5]](https://www.washingtonpost.com/technology/2026/01/27/anthropic-ai-scan-destroy-books/)\n\nThe filings also identify works by the plaintiff authors that appeared in Anthropic’s scanned collection. They include Andrea Bartz’s “The Herd,” “The Lost Night,” “We Were Never Here” and “The Spare Room”; Charles Graeber’s “The Breakthrough” and “The Good Nurse”; and Kirk Wallace Johnson’s “To Be a Friend Is Fatal,” “The Feather Thief” and “The Fishermen.”[[1]](https://www.fbm.com/content/uploads/2025/07/6-23-25-Alsup-Anthropic-Order.pdf)\n\nNot every scanned book was necessarily used to train a commercial Claude model. The court found that Anthropic selected different sets and subsets from its library for different training runs, while retaining the larger collection for research and possible future use.[[1]](https://www.fbm.com/content/uploads/2025/07/6-23-25-Alsup-Anthropic-Order.pdf)\n\n## The new purchasing reports are less conclusive\n\nISBNdb openly presents physical books as an input for AI development. Its public LinkedIn account describes a “compliant path” for acquiring training data through physical books. Its website, as documented by 404 Media, argues that books offer edited and domain-specific knowledge unavailable from ordinary web crawls.[[2]](https://www.404media.co/ai-companies-are-buying-tons-of-old-books-because-theyre-free-of-ai-slop/)[[6]](https://www.linkedin.com/posts/isbndb_a-recent-legal-decision-has-introduced-a-activity-7442851918969954305-3i65)\n\n404 Media also reported that ISBNdb offers confidentiality agreements and acknowledges what it calls the “optics problem” surrounding the destruction of books. Those statements show that the company anticipates destructive scanning among the possible uses of the books it sources. They do not prove that each client destroys its purchases.[[2]](https://www.404media.co/ai-companies-are-buying-tons-of-old-books-because-theyre-free-of-ai-slop/)\n\nAn unnamed bookseller told 404 Media that weekly sales rose from fewer than 20 books to several hundred beginning in April. The orders reportedly lacked a clear pattern by author or subject but consistently involved books with ISBNs. The seller suspected AI-related purchasing but could not identify the buyers or establish their intended use.[[2]](https://www.404media.co/ai-companies-are-buying-tons-of-old-books-because-theyre-free-of-ai-slop/)\n\nOther bookseller accounts demonstrate why the purchasing pattern alone is insufficient. Houston bookseller Charlie Becker documented a 70-book order and demand for obscure titles such as “The Insider’s Guide to Metro Denver” and an old Corel WordPerfect manual. After comparing prices and availability across platforms, Becker concluded that the buyer in his case was probably exploiting resale-price differences—not acquiring AI training material.[[3]](https://charliebecker.substack.com/p/is-an-ai-company-buying-up-all-the)\n\nThe evidence is weaker still for claims involving rare or irreplaceable books. Sellers have expressed concern that low-circulation and out-of-print works could enter destructive scanning pipelines, but the reviewed documentation does not identify a verified rare title that an AI company purchased and destroyed. Nor does it prove that unidentified buyers are attempting to remove the last available copies from circulation.\n\n## The named companies have responded differently\n\nBetter World Books and World of Books did not respond to The Washington Post’s requests for comment about Anthropic’s purchases. Anthropic had also considered approaching New York’s Strand bookstore and public libraries, but the Strand told the Post that it ultimately sold no books to the company. It remains unclear whether any library participated.[[5]](https://www.washingtonpost.com/technology/2026/01/27/anthropic-ai-scan-destroy-books/)\n\nZoom Books, a Canadian company accused in separate European reports of buying out-of-print books for AI use, denied the allegations. It told Publishers Lunch in May that claims it was supplying books for AI scraping were “categorically false” and said it does not digitize or destroy books for AI training or any other purpose.[[7]](https://lunch.publishersmarketplace.com/2026/05/zoom-books-responds-to-accusations-of-selling-books-for-ai-scraping/)\n\nIngram Content Group, a major book distributor, has introduced an opt-out form for publishers that do not want it knowingly selling their products directly to AI companies for model training. Ingram cautions that it may not know the purpose of every transaction but says it will try to honor publisher preferences.[[8]](https://lp.ingramcontent.com/publishers-request/ai-opt-out)\n\n## Buying a book changes—but does not settle—the legal question\n\nIn June 2025, U.S. District Judge William Alsup found that Anthropic’s conversion of lawfully purchased print books into internal digital replacements qualified as fair use on the specific record before him. Each physical copy was discarded, the format change did not increase the number of library copies, and the digital replacements were not distributed outside Anthropic.[[1]](https://www.fbm.com/content/uploads/2025/07/6-23-25-Alsup-Anthropic-Order.pdf)\n\nAlsup also found that using books to train Anthropic’s language models was transformative. The ruling is a federal district-court decision, not an appellate judgment binding courts nationally. Because the parties later settled the remaining dispute, its conclusions were not reviewed on appeal.[[1]](https://www.fbm.com/content/uploads/2025/07/6-23-25-Alsup-Anthropic-Order.pdf)[[4]](https://law.justia.com/cases/federal/district-courts/california/candce/4%3A2024cv05417/434709/680/)\n\nThe ruling did not excuse Anthropic’s separate acquisition of pirated books. The court found that Anthropic had downloaded more than 7 million copies from Books3, Library Genesis and Pirate Library Mirror and rejected fair use as a defense for building a permanent general-purpose library from those pirated files.[[1]](https://www.fbm.com/content/uploads/2025/07/6-23-25-Alsup-Anthropic-Order.pdf)\n\nOn July 20, 2026, a different federal judge granted final approval to a $1.5 billion settlement covering eligible works from the LibGen and Pirate Library Mirror collections. The settlement class covers 482,460 works, at least 91.3% of which had been claimed by rights holders. The agreement resolves past piracy claims; it does not impose damages for Anthropic’s destruction of lawfully purchased print copies.[[4]](https://law.justia.com/cases/federal/district-courts/california/candce/4%3A2024cv05417/434709/680/)\n\nThe available documentation therefore supports a narrower conclusion than the viral headline. Anthropic ran a confirmed industrial program that destroyed millions of purchased books after scanning them. ISBNdb now markets a bulk-acquisition service to other AI developers. The identities of those developers, their total purchases, the titles involved and whether they destroyed the books remain undisclosed.\n\n## Companies mentioned\n\n## Further sources\n\n[[1] U.S. District Court for the Northern District of California, “Order on Fair Use… ↗](https://www.fbm.com/content/uploads/2025/07/6-23-25-Alsup-Anthropic-Order.pdf)\n\n[[2] 404 Media, “AI Companies Are Buying Tons of Old Books Because They're Free of A… ↗](https://www.404media.co/ai-companies-are-buying-tons-of-old-books-because-theyre-free-of-ai-slop/)\n\n[[3] Charlie Becker, “Is an AI company buying up all the used books? (I don’t think … ↗](https://charliebecker.substack.com/p/is-an-ai-company-buying-up-all-the)\n\n[[4] U.S. District Court for the Northern District of California, final settlement a… ↗](https://law.justia.com/cases/federal/district-courts/california/candce/4%3A2024cv05417/434709/680/)\n\n[[5] The Washington Post, “Inside an AI start-up’s plan to scan and dispose of milli… ↗](https://www.washingtonpost.com/technology/2026/01/27/anthropic-ai-scan-destroy-books/)\n\n[[6] ISBNdb, “Compliant AI Training Data via Physical Books: New Legal Path,” compan… ↗](https://www.linkedin.com/posts/isbndb_a-recent-legal-decision-has-introduced-a-activity-7442851918969954305-3i65)+2 more\n\nThe stories that matter, in one email. Free — unsubscribe anytime.", "url": "https://wpnews.pro/news/what-the-evidence-actually-shows-about-ai-companies-destroying-books", "canonical_source": "https://mlq.ai/news/what-the-evidence-actually-shows-about-ai-companies-destroying-books/", "published_at": "2026-07-29 11:52:10.307478+00:00", "updated_at": "2026-07-29 11:52:13.126089+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-ethics", "ai-policy"], "entities": ["Anthropic", "Claude", "Tom Turvey", "Google", "Better World Books", "World of Books", "404 Media", "ISBNdb"], "alternates": {"html": "https://wpnews.pro/news/what-the-evidence-actually-shows-about-ai-companies-destroying-books", "markdown": "https://wpnews.pro/news/what-the-evidence-actually-shows-about-ai-companies-destroying-books.md", "text": "https://wpnews.pro/news/what-the-evidence-actually-shows-about-ai-companies-destroying-books.txt", "jsonld": "https://wpnews.pro/news/what-the-evidence-actually-shows-about-ai-companies-destroying-books.jsonld"}}