For thousands of years, books have been among humanity’s greatest instruments for preserving and transmitting knowledge. They have carried the ideas of one generation to another, the experiences of one civilisation to another, and the discoveries of one scholar to researchers yet to come. A book, therefore, is much more than a collection of words printed on paper. It is a physical record of human thought, memory and civilisation. This is why reports of old and rare books being purchased in large numbers, scanned for artificial intelligence training and then destroyed raise a question that goes far beyond the fate of a few old books: should the physical repositories of human knowledge be sacrificed in the pursuit of technological progress?

Reports from the United States suggest that the demand for old and rare books has increased sharply, with some unidentified buyers reportedly purchasing large quantities, removing their bindings, scanning the pages at high speed and subsequently discarding the physical copies. The issue has received particular attention following disclosures connected with legal proceedings involving Anthropic and reports about a project described as “Project Panama”, under which large numbers of books were reportedly purchased, dismantled and scanned. Reports have also referred to an internal ambition to undertake the destructive scanning of books on an enormous scale. These developments, however, should not be presented as evidence that AI itself has an intention to destroy books. AI has no independent desire or motive of that kind. Such decisions are made by people and organisations seeking large volumes of high-quality human-generated material to train increasingly powerful AI systems.

The need for training data is understandable. But technological necessity cannot by itself settle the ethical question. Does the requirement for data justify destroying the physical forms in which human knowledge has survived? In most cases, it should not. A scan can preserve the words, sentences, chapters and illustrations of a book, but it cannot necessarily preserve everything that makes the original book historically valuable. The paper, typography, binding, cover, publisher’s marks, author’s signature, handwritten annotations, ownership stamps, dedications and even the physical marks left by time can provide important evidence to historians, librarians and researchers. A digital copy may preserve the content of a book; the original can also preserve its provenance and history.

This is not an argument against digitisation. On the contrary, digitisation is one of the most valuable developments in the preservation and democratisation of knowledge. A rare book kept in a library in one part of the world can be made available to students and researchers thousands of kilometres away. Institutions such as UNESCO have recognised digitisation as an important means of preserving documentary heritage and expanding access to it. But digital preservation itself is not permanent. Hardware becomes obsolete, software and file formats change, storage systems can fail, organisations can disappear and ownership and access policies can change. A digital file stored on a private server is therefore not necessarily safer or more permanent than a physical book preserved in a well-managed library or archive.

This creates a deeper concern. If physical copies are systematically destroyed after being digitised, a significant part of humanity’s intellectual heritage could eventually survive only in digital repositories controlled by a small number of private organisations. That would transform the question of preservation into a question of ownership and public access. Human knowledge has historically developed as a cumulative and largely shared enterprise: one scientist builds on another’s research, one writer draws inspiration from earlier literature, and one generation learns from the records preserved by previous generations. Libraries and archives have kept this chain of knowledge alive. If that heritage becomes dependent on privately controlled digital systems, society may eventually have to ask not merely whether the knowledge has been preserved, but who controls access to it.

This is why AI companies have a responsibility that extends beyond the immediate requirements of model training. They must ensure that humanity’s accumulated knowledge is not simply absorbed into private technological systems while the physical and public repositories that preserved it are allowed to disappear. AI has the potential to democratise knowledge on an unprecedented scale. It can make rare books searchable, help translate historical works, connect researchers with archives and bring knowledge previously restricted to particular institutions to people across the world. But that promise will be meaningful only if AI expands access to knowledge rather than concentrating control over it.

The argument that dismantling books is faster and cheaper also deserves careful consideration. Commercial efficiency matters, particularly when millions of pages have to be processed. But when the material involved is rare, historically significant or difficult to replace, speed and cost cannot be the only considerations. Technologies exist that can scan many books without destroying them. Where a physical copy has unique historical or cultural value, preservation should take precedence over convenience. There is also an important distinction between what is legally permissible and what is ethically desirable. A person may have the legal right to dispose of an ordinary book that he or she owns. But a rare book, a unique manuscript or a historically significant document may possess a collective value that extends beyond private ownership.

Governments, national libraries, universities, archives, publishers and AI companies should therefore consider developing common ethical standards for large-scale digitisation. The principle need not be complicated: digitise, preserve and provide access, but do not destroy rare or irreplaceable physical knowledge merely to reduce the cost of digitisation. The objective of AI development should not be to make humanity dependent on machines for access to its own intellectual heritage. Machines should help preserve and distribute knowledge; they should not become its gatekeepers.

The issue becomes especially important when we consider the future. Our children and grandchildren may one day ask where the books of the twentieth and twenty-first centuries went. It would be troubling if the answer were that they had all been scanned and destroyed, leaving their contents dependent on servers operated by corporations whose ownership, business models and access policies could change. Human beings should not outsource the preservation of collective memory entirely to machines or private platforms. Digital copies can give books a second life, but destroying the originals can sometimes mean losing their material history, provenance and physical connection with the past.

Books are not merely paper. They are records of human memory, evidence of human thought and physical traces of civilisation. AI will undoubtedly play an important role in shaping the future, and using thousands of years of accumulated human knowledge to develop it may well be a form of technological progress. But if that progress requires the destruction of the very objects through which that knowledge survived, society must pause and ask whether something valuable is being lost in the name of efficiency.

The principle should therefore be simple: scan and preserve; do not scan and destroy. AI should not become a technology that consumes human knowledge while erasing its physical memory. It should become a technology that helps humanity preserve, understand and share that knowledge more widely. The question is not whether AI should learn from books. It is whether, while teaching machines about humanity’s past, we are prepared to destroy the objects that preserved that past in the first place. That is not a question for AI to answer. It is a question for human society.