AI Companies Purchase and Destroy Millions of Books for Training Data

AI companies are buying millions of used, rare, and out-of-print books in bulk to train their language models, then destroying the physical copies after scanning their contents. This practice, known as destructive scanning, has become widespread as developers seek high-quality training data written entirely by humans before generative AI became common. Books published before 2022 or 2023 are especially valuable because they provide "clean" data uncontaminated by AI-generated text.

A federal court ruling in June 2025 upheld this practice as fair use under copyright law. Judge William Alsup in San Francisco determined that buying physical books, dismantling their spines, scanning the pages, and then recycling the remains qualifies as transformative use, provided the companies own the books and keep the digital copies for internal use without redistributing them. Similar rulings have since been issued in cases involving OpenAI and Meta. However, a separate federal case approved a $1.5 billion copyright settlement requiring Anthropic to pay thousands of authors about $3,000 per book after the company used pirated copies to train its AI model, Claude. These conflicting outcomes highlight ongoing legal disputes over how AI companies obtain and use copyrighted material.

The demand for books has reshaped the used-book market. Some sellers report weekly sales jumping from around 20 books to several hundred. Orders often appear random, spanning multiple topics and ignoring normal pricing patterns. Buyers frequently remain anonymous, with transactions covered by nondisclosure agreements.

A company called ISBNdb has played a key role in this market, offering to source large volumes of physical books for AI training. The company previously advertised orders ranging from 1,000 to 1 million titles, emphasizing confidentiality. While ISBNdb has since removed its AI book-sourcing page—claiming it was only testing demand—similar services continue to operate. Booksellers on platforms like Alibris and Biblio report a sharp increase in bulk orders, though buyers often remain unidentified.

The process typically involves destructive scanning, where book bindings are removed or sliced off to feed pages into high-speed industrial scanners. Video footage from a Canadian digitization company showed this process in action. Some AI developers, including Elon Musk, have publicly opposed the destruction of books; Musk stated that rare books should be preserved in libraries and scanned carefully rather than dismantled. Internal documents from Anthropic reportedly acknowledged ethical concerns, with one memo stating the company did not want the public to know about its book-destruction practices.

Booksellers express mixed reactions. While some benefit financially from clearing old inventory, others express discomfort about the destruction of rare or uncommon books. The trend has raised concerns about the permanent loss of out-of-print titles that may never be preserved or made available to the public again. Similar patterns have been observed internationally, with rare booksellers in Europe reporting large requests for thousands of English-language titles, often from companies suspected to be AI firms.

The long-term impact on book preservation, copyright law, and public trust in AI development remains uncertain. Proponents argue the practice completes a book’s lifecycle without depriving authors of additional income, while critics compare it to book burning and warn of cultural loss. The debate continues as AI companies prioritize access to human-authored training data.

Original Sources/Tags: techspot.com, tomshardware.com, dallasexpress.com, yahoo.com, futurism.com, finance.yahoo.com, russh.com, x.com, (anthropic)

This article provides almost no actionable help to a normal reader. It describes a market trend where AI companies purchase and destroy books for training data, but it offers no clear steps, choices, or tools a person can use. There are no instructions on how to participate in this market as a seller, no guidance on how to protect rare books, and no resources for authors or publishers concerned about their work. The only implied action—observing the trend—is passive and does not empower the reader to take meaningful steps. The article does not link to organizations that preserve rare books, legal resources for creators, or platforms where sellers might verify bulk buyers. Without these, the reader is left with nothing to do but read and move on.

The educational depth is limited. While the article explains the mechanics of the market—such as bulk purchases by ISBNdb and destructive scanning—it remains superficial. It does not explore why older books are preferred for AI training, how destructive scanning compares to non-destructive methods, or what legal precedents exist beyond the single fair use ruling mentioned. The article mentions that books provide "cleaner" data than web content but does not explain why this matters or how it affects AI model performance. The numbers, such as "1,000 to 1 million titles," are presented without context about how these volumes compare to typical AI training datasets or what percentage of the market they represent. The lack of explanation leaves the reader without a deeper understanding of the trend’s significance.

The personal relevance is narrow. The article affects booksellers, authors, and publishers who may interact with this market, but it does not explain how these groups can protect their interests. For booksellers, it does not clarify how to identify legitimate bulk buyers or negotiate fair prices. For authors, it does not address how to track the use of their work or seek compensation. For most readers, the relevance is indirect: the trend may influence the availability of older books or the quality of AI models, but the article does not connect these outcomes to the reader’s own life. It does not address how someone might assess the rarity of their books, evaluate offers from bulk buyers, or decide whether to sell. The lack of practical connection makes the issue feel distant and abstract.

The public service function is minimal. The article raises awareness of a market trend but does not provide warnings, safety guidance, or emergency information. It does not explain how rare books might be preserved, how creators can protect their work, or how the public can advocate for ethical AI training practices. The article functions as a news report rather than a public service resource, offering no tools to help readers act responsibly or make informed decisions. It does not even clarify whether the destruction of books is legal in all jurisdictions or how sellers might verify the legitimacy of bulk buyers.

The practical advice is nonexistent. The article mentions that booksellers have mixed feelings about the trend but does not suggest how they might navigate these concerns. There are no tips on how to document book sales, how to research bulk buyers, or how to advocate for ethical AI training practices. The guidance is so vague that it is effectively useless. For example, the article notes that orders appear "random" and ignore pricing patterns, but it does not explain how sellers can assess whether an offer is fair or how to avoid contributing to the destruction of rare books.

The long-term impact is unclear. The article focuses on the immediate market trend and its implications for booksellers and AI companies, but it does not help readers plan ahead . It does not explain how to evaluate the long-term risks of selling books for AI training, how to preserve rare or low-circulation books, or how to prepare for potential changes in copyright law. The lack of forward-looking guidance means the reader gains no lasting benefit . The article does little to help someone understand how to navigate future conflicts between AI development and book preservation.

The emotional and psychological impact leans toward unease without empowerment. The article describes the destruction of books—a practice that may unsettle readers who value physical media—but offers no constructive way to respond. The focus on the loss of rare books and the secrecy of transactions may leave readers feeling anxious or powerless, but without any tools to address these concerns. The tone is more alarming than reassuring, as it highlights risks without providing solutions. The article does note that proponents argue the practice completes a book’s lifecycle, but this perspective is presented without depth or context, leaving readers without a balanced view.

The language avoids overt clickbait, though the headline and opening sentence emphasize the scale of the trend to capture attention. The framing is factual but prioritizes engagement over substance. The article does not sensationalize the issue, but it also does not provide enough depth or practical guidance to make the story truly useful. It relies on dramatic details, such as the destruction of books and the secrecy of transactions, to maintain interest without offering meaningful insights.

The biggest missed opportunity is failing to explain how ordinary people can engage with or respond to this trend. The article presents a problem—the destruction of books for AI training—but offers no tools to address it. It could have included basic steps for booksellers to verify bulk buyers, for authors to track the use of their work, or for readers to preserve rare books. Instead, the reader is left with a story and no way to act on it.

If you are a bookseller, author, or someone who values physical books, here are some universal steps you can take to navigate this trend. These principles apply regardless of your specific role or the types of books you handle.

Start by understanding the market. If you are a bookseller, recognize that bulk buyers may approach you with offers to purchase large quantities of books. These buyers often prioritize older titles and may not disclose their ultimate use of the books. Before accepting an offer, research the buyer’s reputation. Look for reviews or feedback from other sellers, and ask for references if possible. If something feels off—such as an unusually high offer for low-value books or a lack of transparency—proceed with caution. Trust your instincts and avoid deals that seem too good to be true.

Document your sales. Keep records of all transactions, including the buyer’s name, contact information, and the titles sold. This documentation can be useful if questions arise later about the books’ fate or if you need to verify the legitimacy of the sale . If you are an author or publisher, track where your books are being sold and who is purchasing them. While you may not have control over how buyers use your books, awareness can help you make informed decisions about future publications or distribution strategies.

Evaluate the rarity and value of your books. Before selling, assess whether the books you are offering are rare, out of print, or culturally significant. If they are, consider whether their destruction would be a loss to the broader community. Some books may have value beyond their content, such as historical significance or limited availability. If you are unsure, consult with librarians, archivists,or rare book dealers who can provide guidance . Preserving rare books can be just as important as selling them, especially if they are unlikely to be reprinted or digitized.

Explore alternatives to selling. If you are uncomfortable with the idea of your books being destroyed for AI training, consider other options. Donate books to libraries, schools, or community organizations that can put them to use. Some institutions may even offer tax benefits for donations. You can also sell individual copies to collectors or readers who value physical books. While these options may not be as lucrative as bulk sales, they can help ensure your books remain accessible to future readers.

Stay informed about legal and ethical developments. AI training practices and copyright laws are evolving rapidly. Follow updates from organizations that advocate for creators’ rights, such as the Authors Guild or the American Library Association. These groups often provide resources and guidance on how new technologies affect books and publishing. Understanding the legal landscape can help you make informed decisions about selling or preserving your books.

Advocate for ethical AI practices. If you are concerned about using books for AI training, consider supporting organizations that promote ethical AI development. These groups may advocate for transparency in training data, fair compensation for creators, or the preservation of cultural works. Collective action can be more effective than individual efforts, especially in shaping industry standards. You can also engage with policymakers to encourage regulations that balance innovation with the protection of creative works.

Prepare for the future. The market for books in AI training is likely to grow, so stay adaptable. If new opportunities or risks emerge, take the time to understand how they might affect you. For example, if copyright laws change to allow broader use of books for AI training, consider how this might impact your rights as an author or seller. If preservation efforts gain traction, explore ways to contribute to or benefit from them. Being proactive can help you navigate changes in the market more effectively.

Finally, prioritize what matters to you. Whether you are a bookseller, author, or reader, your values should guide your decisions. If preserving books is important to you, seek out ways to protect them. If financial gain is your priority, weigh the risks and benefits of bulk sales carefully. There is no one-size-fits-all answer, but staying true to your principles can help you make choices you feel good about. The book market is changing, but your agency in how you engage with it remains intact.

The text says "to avoid contamination from AI-generated text, which can degrade model performance over time." This makes AI-generated text sound like a disease or poison. The word "contamination" is strong and makes readers feel that AI text is bad and dirty. It helps AI companies by making old books seem like the only clean choice. The order puts the problem first, so readers think AI text is always bad.

The text says "cleaner, more authoritative data than web content." The word "cleaner" makes web content sound messy or wrong. The word "authoritative" makes books sound like they are always right. This hides that books can be wrong or old. It helps AI companies by making books seem better than other data. The order puts books first, so readers think books are the best choice.

The text says "mixed feelings, benefiting financially while disliking the end use and the destruction of uncommon books." The words "mixed feelings" make the bookseller sound fair and thoughtful. This hides that the bookseller still sells books to be destroyed. It helps the bookseller look like they care, even if they still do the sale. The order puts the good feeling first, so readers remember that more.

The text says "completes a book’s lifecycle without depriving creators of additional income." The words "completes a lifecycle" make destroying books sound natural and good. This hides that the books are gone forever. It helps AI companies by making destruction seem like a normal part of a book’s life. The order puts the good part first, so readers think it is fair.

The text says "critics highlight concerns about the loss of rare or low-circulation books." The word "critics" makes the people who care about books sound like they just complain. This hides that their worry might be real and important. It helps AI companies by making critics seem weak or picky. The order puts critics after the AI company view, so readers think the AI view is more important.

The phrase "destructive scanning" is used but not explained until later. The word "destructive" sounds bad, but the reader does not know why. This makes the process seem worse than it is until the text says it is fair use. It helps critics by making the process sound bad first. The order hides the fair use part until after the bad word.

The phrase "federal judge recently ruled that this practice qualified as fair use" uses passive voice. The words do not say who did the ruling or why. This hides that a judge made the decision, not the law itself. It helps AI companies by making fair use sound like a fact, not a court choice. The order puts the ruling after the bad word, so readers remember the bad word more.

The text says "proponents argue the practice completes a book’s lifecycle." The word "proponents" makes the people who support destruction sound like they have a real reason. This hides that their reason might be weak. It helps AI companies by making their view sound thoughtful. The order puts proponents after critics, so readers see both sides but think the AI side is more important.

The phrase "high-quality, human-written content" makes books sound better than other data. The words "high-quality" and "human-written" make books seem special. This hides that other data can be good too if it is checked. It helps AI companies by making only books seem useful. The order puts books first, so readers think books are the best choice.

The text says sellers report "a sharp increase in bulk orders, though buyers remain unidentified." The words "remain unidentified" make buyers sound secret and maybe bad. This hides that buyers might just want privacy. It helps critics by making buyers seem sneaky. The order puts the secret part last, so readers remember it more.

The text conveys several meaningful emotions that shape how readers perceive the growing market for books used in AI training. One of the strongest emotions is **unease**, which appears in phrases like "destruction of uncommon books" and "loss of rare or low-circulation books." These words suggest that something valuable is being lost, making readers feel uncomfortable about the practice. The unease is reinforced by the mention of "mixed feelings" from a bookseller who benefits financially but dislikes the end use, which adds a human element to the concern. This emotion serves to make readers question whether the destruction of books is justified, even if it helps AI companies. It positions the trend as something that may harm cultural heritage, steering the reader toward skepticism rather than acceptance.

Another key emotion is **distrust**, which surfaces in the discussion of confidentiality and unidentified buyers. Phrases like "buyers remain unidentified" and "every transaction covered by nondisclosure agreements" create a sense that the market operates in secrecy, making readers suspicious of the motives behind these purchases. The distrust is further amplified by the mention of orders that "appear random" and ignore normal pricing patterns, which suggests that the buyers may not be acting in good faith. This emotion serves to make readers wary of the AI companies and the middlemen like ISBNdb, framing them as entities that prioritize their own interests over transparency or ethical concerns. The effect is to make the reader more critical of the entire practice, rather than seeing it as a straightforward business transaction.

A sense of **relief** appears briefly but is overshadowed by the other emotions. It emerges in the mention of the federal judge’s ruling that destructive scanning qualifies as fair use, which could make some readers feel reassured that the practice is legally permissible. The relief is mild, however, because it is immediately followed by the acknowledgment that critics still have concerns. This emotion serves to provide a counterbalance to the unease and distrust, suggesting that the practice is not entirely without justification. However, the relief is weak compared to the other emotions, so it does little to shift the overall tone of the message. Instead, it functions as a brief acknowledgment that there are legal arguments in favor of the trend, without fully endorsing them.

The text also conveys a subtle sense of **indifference** from the perspective of the AI companies and proponents of the practice. Phrases like "completes a book’s lifecycle without depriving creators of additional income" and "high-quality, human-written content" make the destruction of books sound like a natural and unproblematic part of progress. This indifference is not an emotion felt by the reader but rather an attitude attributed to the proponents, which serves to highlight the contrast between their perspective and the concerns of critics. By presenting the proponents as unconcerned with the loss of books, this attitude reinforces the unease and distrust felt by the reader, making the practice seem even more questionable.

The emotions in the text work together to guide the reader toward viewing the trend with skepticism. The unease and distrust create a sense of moral discomfort, making the reader question whether destroying books for AI training is ethically acceptable. The brief relief provided by the legal ruling does little to counter these emotions, as it feels more like a technicality than a full endorsement. The indifference attributed to the proponents further alienates the reader, making the trend seem cold and calculating. Together, these emotions shape the message to make the reader more likely to side with the critics and view the practice as harmful to cultural heritage, even if it benefits AI development.

The writer uses emotional language strategically to amplify the concerns about the trend. Words like "destruction" and "loss" are stronger than neutral terms like "use" or "removal," making the practice sound more extreme and harmful. Repeating the idea of secrecy—through phrases like "unidentified buyers" and nondisclosure agreements—reinforces distrust and makes the market seem shady. The personal story of the bookseller with "mixed feelings" adds a human touch, making unease feel more relatable. The writer also contrasts the perspectives of critics and proponents, framing the latter as indifferent to the cultural value of books while emphasizing the former’s moral concerns. This contrast makes the reader more likely to align with the critics. By choosing words that sound emotional rather than neutral, the writer steers the reader’s attention toward seeing the trend as problematic, rather than a simple business practice. The emotional tools—strong language, repetition , personal stories, and contrast—work together to shape the reader’s reaction and make the message more persuasive.