AI firms are quietly buying and destroying millions of printed books to train their models
TL;DR AI
2 min readKey summary
AI companies are increasingly buying bulk lots of older printed books, especially pre-2022 titles, to use as training data for models.
Book-sourcing services and sellers like ISBNdb, Alibris, and Biblio say demand has risen for large-scale, confidential acquisitions.
The report says some books are scanned destructively, and Anthropic court filings indicate millions of purchased books were scanned while many copies were destroyed.
The trend raises concerns about transparency, copyright, and the loss of rare or unwanted books in the training-data pipeline.