Current Affairs Security

Why Copyright Law Just Gave AI Firms a Licence to Shred Your Books

2026-07-29

In June 2025, U.S. District Judge William Alsup ruled that digitising legally purchased print books and using them to train large language models was fair use. Sensible enough, maybe. Except the logic he built in has since become a blueprint for industrial-scale book destruction. The judge noted that the process replaced print copies with digital versions without adding new copies. That reasoning has now evolved into something rather darker: a legal template that actively rewards companies for burning the originals.

Anthropic got there first. Under an internal operation called Project Panama, the company hired Tom Turvey (formerly Google Books' head of partnerships) to source "all the books in the world." They cut up millions of printed books, scanned them, and discarded the originals. Court filings unsealed in connection with a copyright lawsuit show the vendor proposal referenced converting anywhere from 500,000 to two million books over six months. The company later settled related claims involving pirated digital books for $1.5 billion. The ruling on purchased books, though? Still stands.

What started as one company's internal supply chain has now gone public and commercial. ISBNdb, a database broker, is openly marketing its services as a middleman for AI developers who need physical books. They now facilitate orders ranging from 1,000 to one million books at a time. The company even describes itself as operating the world's largest book database. It's repositioned its business squarely towards feeding the AI training machine.

Pre-2022 publications are the prize. They predate the large language model era and carry no AI-generated contamination. ISBNdb marketing makes this explicit: books published before widespread LLM adoption are "structurally guaranteed to be free of this contamination." There's a reason they're premium goods. Scanning at scale usually destroys the book anyway—workers slice off the spine so loose pages feed through a machine. Faster. Cheaper. No careful handling required.

ISBNdb openly acknowledges the optics problem. Their site states plainly: "The optics problem is real. 'AI company destroys two million books' is not a headline that generates sympathy." The workaround they suggest is simple. Reword the deed as digital preservation. Hide the buyer names under strict NDAs. Problem solved.

Used-book dealers have begun noticing. One anonymous professional seller told 404 Media about an unprecedented surge beginning in April 2026. They used to shift around 20 books in a good week. Recent months? Several hundred per week. That's a fivefold increase over baseline. Dealers are expressing concern about uncommon editions—rare volumes with limited surviving copies being cleared from circulation entirely.

The copyright logic appears deliberately constructed. Judge Alsup held that converting lawfully purchased print books into non-distributed digital copies was fair use because the PDFs replaced the purchased books without increasing the library's copy count. Destruction did practical legal work. Without it, both book and scan coexist—a fact pattern the court's one-for-one logic made legally inconvenient.

No title-level evidence yet documents a specific rare book being destroyed. Anecdotal reports circulate among dealers and on social media. Nothing definitive in public records. What is certain is this: the legal framework now rewards wholesale acquisition and destruction rather than preservation or circulation. Copyright law has quietly shifted its valuation of the physical record.

Source & further reading:

Sources