História
agosto 17, 2026
Caça da Amazon por Dados de IA Coloca Livros Raros na Linha de Fogo
Relatos que ligam compras em massa de livros e digitalização destrutiva ao treinamento de IA ampliaram temores de que volumes insubstituíveis estejam sendo sacrificados em nome da escala. A Amazon afirma que compra livros por canais comerciais, enquanto defensores da preservação argumentam que a corrida tecnológica não exige destruí-los.
The race to feed artificial intelligence ever more text is colliding with a simpler, older imperative: some books cannot be replaced once they are cut apart.
The alarm intensified after 404 Media placed a tracker in a rare book and reported that it ultimately reached Amazon’s VGT3 facility in Las Vegas. The report described Amazon-linked purchases of books whose spines were removed so their pages could be scanned for model training. Amazon did not address the destruction allegation directly, saying only that it “purchases books through commercial channels to improve the products and services customers use.”1
The commercial logic is blunt. Large language models require huge quantities of reliable, long-form text, and out-of-print books can contain material unavailable online — including text written before AI-generated prose began to pollute the data supply. But booksellers and preservation advocates see a cost that does not fit neatly into a training pipeline: a rare physical copy, once dismantled, is gone.
Concern had already been building since revelations that Anthropic destroyed millions of print books for training, though there is no indication it destroyed rare volumes. Reports of suspicious bulk orders have since made the fear more concrete. Kennys Bookshop in Ireland recently flagged a buyer seeking 5,000 obscure titles without haggling — behavior booksellers increasingly treat as a warning sign of AI sourcing.2
The dispute is not over whether books can be digitized, but how. Google patented non-destructive scanning technology years ago, while the Internet Archive uses trained workers to turn fragile pages by hand. Its scanners concluded that automated systems were ill-suited to brittle books and special collections; “clean, dry human hands are the best way to turn pages,” archive staff said.2
AI companies and their defenders can argue that purchased copies may legally be scanned. Critics counter that legality is a poor answer to cultural loss. As Kennys’ Tomás Kenny put it, if a buyer is scanning a book “with a view to taking intellectual property,” his shop does not want to participate.2