Amazon, which started off selling books, is now destroying rare texts to train AI. That sentence sounds like a headline from a cautionary tale, yet here we are. The logic is grimly efficient: large language models have already absorbed whatever is freely available online, so the next frontier of training data lies in the physical archives, the fragile, one-of-a-kind volumes that never made it to the digital commons. And if that means cutting the spines, scanning the pages, and discarding the originals, so be it. The cost of progress, it seems, is measured in the very artifacts we built to preserve knowledge.
Let's be direct about what this means for you, the person who relies on spreadsheets and data models to make sense of the world. If the people building AI are willing to destroy a rare book to extract its text, they are making a calculated bet that the model's capabilities matter more than the source material's existence. That is not a neutral trade. It is a statement about what we value, and it should make you pause the next time you ask an AI to summarize a report or draft an email. The same technology that promises to simplify your workflow is built on a foundation that treats physical knowledge as disposable. This is not an argument against innovation; it is an argument for understanding the cost of the convenience you are being offered.
We have written before about the strange, unsettling experience of interacting with AI clones, and about the practical mechanics of training models at scale. The through-line is that AI is only as good as the data it consumes, and the data it consumes is increasingly scarce. That scarcity is driving companies to extreme measures, and the rare-book trade is just the latest casualty. But here is the uncomfortable question: if a company will destroy a book to train its model, what will it do to the people who contributed that data once it becomes a liability? The same logic that justifies the destruction of a physical object applies to the erosion of consent, privacy, and authorship. You are not just a user of these tools; you are a data point in a system that has already decided your contributions are expendable.
So what do we tell a reader who asks, "Should I stop using AI?" That is the wrong question. The right question is: what are you willing to lose? The answer is not to abandon the technology, but to demand transparency about where the training data comes from and what is sacrificed to get it. If a company cannot tell you whether it destroyed a book to build its model, it has no business asking for your trust. The concrete point to watch is this: as rare texts become harder to find, the value of those that remain will only increase, and the pressure to digitize, copy, or simply erase them will grow. The next time you see a headline about a library selling off its archives, ask yourself who is buying and why. The answer might be a machine that will never read the book, only mine it for patterns, leaving the rest of us to wonder what we lost in the process.