
Artificial intelligence (AI) companies are purchasing millions of physical books, especially rare and out-of-print titles, spanning every page and destroying originals to train their large language models.
Anthropic, the Google-backed company behind the Claude AI chatbot, led the effort under a secret internal program codenamed “Project Panama.”
Industry reports reveal that the company spent tens of millions of dollars acquiring books from used-book sellers such as Better World Books and World of Books. Via hydraulic cutting machines, removed their pages before scanning them with industrial-grade equipment.
The physical copies were then discarded or recycled.
According to an internal Anthropic planning document, the company stated: “We don’t want it to be known that we are working on this.”
The strategy involves books published before 2022 since these books hold value as they were produced before the era where AI was generating a large number of books to be uploaded online. Inputting AI-generated text into models will lead to “model collapse” as time goes on.
In June 2025, a ruling by a federal judge declared that scanning and destroying a book after purchasing it is allowed in accordance with the first-sale doctrine, where the buyer has the right to dispose of the item once it has been bought.
Earlier, Anthropic agreed to a $1.5 billion settlement in July over using pirated books to train earlier models in a separate case. Affected authors are offered around $3,000 per title.
Bulk purchasing of rare books not usually collected by bibliophiles is becoming increasingly common throughout Europe. An online company called ISBNdb now assists companies in purchasing books for AI purposes and guarantees confidentiality as well as one million book purchases per order.
ISBNdb states on its website: “The world’s best AI training data is sitting on a shelf.”
The bookshops in Australia have also expressed fear that rare books with very few copies left will be withdrawn from the shelves. According to Elon Musk, the SpaceXAI team would save such books and “scan them the hard way.”
2026-07-30 23:51:00










