Books Destroyed, History at Risk — All for AI Training
The rapid expansion of artificial intelligence is creating an unexpected problem for the world of books: what happens to physical copies when they are acquired primarily to train AI systems?
Court documents have revealed that Anthropic, the company behind Claude, purchased millions of books as part of an effort to build a large digital collection for AI training.
The company used destructive scanning, a process in which books are dismantled so that their pages can be scanned efficiently. After digitisation, the physical copies were discarded.
Anthropic has said that the books were legally acquired and that it did not specifically target rare books for destruction.
The case has nevertheless raised concerns among people involved in preserving historical material.
For ordinary modern books, destroying a physical copy may have little consequence. Older and rarer books are different. Their value may extend beyond the text printed on their pages.
A centuries-old book can contain handwritten annotations, signatures, stamps, illustrations, corrections and evidence of previous ownership. Its paper, binding and printing techniques can also provide information to researchers.
Those details cannot necessarily be reconstructed from a digital copy.
The issue has become more complicated as demand for AI training material has increased. Large AI systems require enormous quantities of text, creating an incentive to acquire and digitise books on a large scale.
Reports in 2026 have also drawn attention to the purchase and processing of rare and out-of-print books for scanning. In some cases, books have reportedly been sent to facilities where they can be dismantled as part of the digitisation process.
It remains important to distinguish between documented destruction of books for digitisation and claims that companies are deliberately destroying historically unique material. Evidence currently available does not establish that Anthropic systematically destroyed rare historical books.
However, preservation specialists face a broader question: whether existing safeguards are sufficient when physical books become raw material for large-scale data collection.
Libraries and archives normally preserve original copies even when high-quality digital versions are available. This is because the physical object can contain information that is not visible in a digital reproduction.
The debate comes as AI companies also face legal disputes over the use of copyrighted books and other works in training datasets.
For the publishing and preservation communities, the challenge is therefore becoming twofold: ensuring that books can be used to advance new technologies while also ensuring that irreplaceable physical records of the past are not unnecessarily lost.
The technology can preserve the words in millions of books.
The question is whether the original books will survive alongside them.
