AI Companies Are Bulk Buying and Destroying Books

Major AI companies have begun mass purchasing physical books to train their models and build new databases. According to an investigation by 404 Media, the books are completely destroyed physically after the digitization process, raising serious concerns about the fate of literary heritage. This is reported by Ixbt.com reports .

It turns out that such purchases are carried out through special intermediaries in order to hide the ultimate clients. As part of some wholesale transactions, editions ranging from 1,000 to 1 million copies are bought at once. Here, buyers are more interested in the volume and quality of the text rather than the author, genre, or specific topic.

Dramatic shifts in the book market

The ISBNdb database, which previously served libraries and bookstores, is also actively participating in this process. Currently, it offers wholesale delivery and bulk purchasing services of books for corporations working in the field of AI. Second-hand book sellers note that demand has surged sharply — whereas previously an average of about 20 books were sold per week, this figure has now reached hundreds of copies. A similar trend is observed on the Alibris and Biblio marketplaces.

According to experts, printed works written by humans are considered the cleanest and highest-quality source of data for AI. The reason is that they lack the volume of repetitive and low-quality data generated by neural networks themselves. This plays an important role in improving the logical structure and literacy of language models.

Legal issues and disappearing heritage

This practice was previously confirmed in court documents filed against Anthropic. The company bought books, scanned them with the help of contractors, and then tore many copies into individual pages to make the fast industrial scanning process easier. The court found using legally purchased books for training to be permissible under the "fair use" doctrine. However, the company has also faced other lawsuits for storing millions of pirated books. Additionally, Google has been sued for illegally using millions of copyrighted books to train its Gemini model.

The most alarming aspect of the process is that after scanning, many rare and long-out-of-print books are irreversibly destroyed. Their digital copies are kept only in closed AI databases. This poses a threat of the physical disappearance of valuable cultural heritage for future generations.