Overview of AI Companies Are Buying and Destroying Used Books
This episode of The Journal investigates a strange surge in bulk used-book orders that led sellers to uncover a likely AI-related book acquisition pipeline. The reporting suggests that companies like Anthropic have been buying physical books in large quantities, scanning them for model training, and then destroying the originals—raising legal, ethical, and economic questions for the used-book world.
What was happening in the used-book market
Used booksellers began noticing unusual order patterns:
- Multiple orders from the same mysterious buyer over a short period
- Large bulk purchases, sometimes 50 to 100 books at a time
- Orders focused on obscure nonfiction titles rather than popular books
- Shipments going to warehouses rather than individual readers
Examples of the kinds of books being bought included highly niche or academic titles like:
- Cultures of Glass Architecture
- Slow Wine 2012
- Women in Television News Revisited
- Marriage Litigation in Medieval England
At first, these orders were a financial boost for sellers, sometimes doubling revenue. But the volume and secrecy soon made the activity suspicious.
How the mystery was uncovered
A bookseller in the Pacific Northwest decided to investigate by placing a tracker inside a book package. Using a slim, credit-card-sized tracking device and removable tape, she followed two books after sale.
Tracker results
- First book: Ohio warehouse → Illinois industrial site tied to book scanning → likely destroyed
- Second book: Wisconsin warehouse → Las Vegas area near Amazon facilities → Mexico at a recycled paper manufacturer
Her conclusion: the books were likely being routed into an AI-related operation that scanned and then destroyed them.
The AI connection: Anthropic and “Project Panama”
The episode connects the mystery orders to court filings in the copyright lawsuit against Anthropic.
Key points
- Anthropic was accused of downloading millions of pirated digital books
- Separate reporting showed the company also bought physical books in bulk
- The internal codename for the physical-book project was Project Panama
- The goal was reportedly to train AI models using information from print books that weren’t easily available online
The episode explains that AI companies already train on web data, but print books contain information that isn’t fully digitized elsewhere.
Why the books were being destroyed
The episode describes two scanning methods:
-
Non-destructive scanning
- Page-by-page scanning
- Too slow for millions of books
-
Destructive scanning
- A machine cuts off the book’s spine
- Pages are scanned in bulk
- Much faster and more efficient for high-volume digitization
Anthropic’s method reportedly relied on the destructive approach. The company said it does not buy and destroy rare or antiquarian books, but the practice still drew backlash.
Legal and ethical fallout
A central irony of the story is that destroying the physical books may have helped Anthropic legally.
Why it mattered
- A judge ruled that buying a physical book, scanning it, and destroying the original could be legal under copyright law in this context
- The logic: the company was replacing the print copy with a digital version for internal use, not adding a new copy to the market
That precedent appears to have encouraged other AI companies to adopt similar practices.
Ethical concerns
- Rare or obscure books could be lost forever
- Multiple AI firms buying the same uncommon title could drain the market
- Booksellers feel conflicted about facilitating destruction of books they love and curate
How booksellers are responding
Booksellers interviewed in the episode were divided:
- Some are uncomfortable helping destroy books, especially if they may be rare or meaningful
- Some accept it pragmatically, noting that books are destroyed in other parts of the industry too
- Some draw a hard line and refuse suspicious orders for rare books
The episode notes that publishers also regularly pulp unsold books, so destruction is not new—what’s new is AI companies buying books specifically for scanning and destruction.
Main takeaways
- AI companies are increasingly sourcing physical books to train models, not just digital text
- Used-book sellers stumbled onto the trend through unusual bulk orders
- A tracking experiment strongly suggested books were being scanned and destroyed
- The practice sits in a gray area that may be legal but still feels ethically troubling
- The immediate impact may be modest for the broader book market, but it could be significant for rare and hard-to-find books
- The story highlights a bigger shift: valuable human knowledge is being converted from print into AI training data, often without readers ever seeing the original books again
Bottom line
The episode presents a fascinating but unsettling collision between the used-book economy and AI development. What began as a business mystery turned into a story about copyright law, data extraction, and the possible disappearance of physical books into machine learning pipelines.