AI Companies Accused of Destroying Rare Books After Scanning Them to Train Models

Artificial intelligence companies are facing growing criticism after reports emerged that some firms have been buying large numbers of physical books, including rare, antique and out-of-print editions, before scanning their contents to train AI models and discarding the original copies.

The practice has sparked concern among booksellers, librarians and cultural preservationists, who argue that while digitising books can help preserve knowledge, destroying physical copies risks erasing historical artefacts that cannot be replaced.

The controversy follows the release of court documents detailing how AI company Anthropic purchased millions of printed books, removed their bindings, scanned every page and disposed of the originals as part of a project to expand the training data for its Claude chatbot.

Although destructive scanning has long been used in some large-scale digitisation projects, critics argue that the growing demand for high-quality training data has accelerated the practice and may now be extending beyond common second-hand books to include increasingly scarce editions.

Reports from Europe and Australia suggest rare book dealers have noticed unusual bulk purchases of obscure titles believed to be destined for AI training.

Why AI Companies Want Physical Books

As AI developers seek new sources of high-quality text, physical books have become increasingly valuable.

Much of the internet is now saturated with AI-generated content, making older printed works especially attractive because they contain professionally edited, human-written material.

Unlike websites that may have been repeatedly copied or altered, books often provide clean, consistent text that improves the quality of AI training datasets.

Court filings in Anthropic's copyright case revealed the company launched an internal initiative known as 'Project Panama', which involved buying books in bulk before removing their bindings so pages could be fed through industrial scanners.

According to the documents, this approach was significantly faster and cheaper than scanning books without damaging them.

Why the Practice Is So Controversial

While digitising books is not unusual, the destruction of physical copies has become the focus of criticism.

Librarians and antiquarian booksellers argue that older books are valuable for more than just the words printed on their pages.

First editions, handwritten notes, ownership inscriptions, original bindings and other physical characteristics can all have historical or cultural significance that cannot be captured through scanning alone.

Several booksellers have reported receiving unusually large requests for obscure titles, leading to fears that increasingly rare books could disappear from the second-hand market.

Some preservation experts have warned that once these copies are destroyed, future researchers lose access to unique physical artefacts, even if digital versions survive.

What Courts Have Said

The legal picture remains complex.

In one of the first major AI copyright decisions, US District Judge William Alsup ruled that Anthropic's digitisation of legally purchased books for internal AI training qualified as fair use because the company bought the books before scanning them and did not redistribute digital copies of the original works.

However, the judge separately criticised Anthropic's earlier use of pirated books, which formed part of ongoing copyright litigation.

The ruling addressed copyright law rather than the ethics of destroying physical books, leaving many cultural organisations concerned that legally permissible practices could still result in the permanent loss of important printed works.

A Growing Debate Over AI and Cultural Heritage

The controversy has fuelled broader questions about how AI companies should balance technological progress with cultural preservation.

Some experts argue that scanning endangered books creates valuable digital archives that can help preserve knowledge for future generations.

Others counter that destroying original copies, particularly rare editions, weakens museums, libraries and private collections that protect literary history.

Industry groups have called for greater transparency over how AI companies obtain training materials and whether more non-destructive scanning methods could be adopted.

While destructive scanning is common for mass-market books with millions of surviving copies, preservationists argue that the same approach should not be applied to rare or historically significant works.

As demand for AI training data continues to grow, the debate is likely to intensify.

Supporters of the practice argue it helps build more capable AI systems using legally acquired material, while critics warn that sacrificing physical books for digital progress could come at an irreversible cultural cost.