Stop AI Companies From Destroying Books for Training Data

Congress needs to act now to stop AI companies from bulk-purchasing and physically destroying books — including rare and potentially irreplaceable copies — to use as training data. This is happening at scale, right now, and there is no law preventing it. Here is what is happening: companies like Anthropic are buying physical books by the millions, cutting them apart with hydraulic machines, scanning the pages, and discarding the remains. They do this specifically to exploit the first-sale doctrine and sidestep copyright law — a loophole a federal judge has already blessed as "fair use." A database service called ISBNdb is actively marketing pre-2022 physical books as premium AI training material and promising anonymity to buyers because, as ISBNdb itself admits, "AI company destroys two million books is not a headline that generates sympathy." Small booksellers report going from selling 20 books a week to hundreds overnight, with rare and out-of-print titles — some possibly among the last surviving copies — disappearing into this pipeline. This is a cultural heritage crisis dressed up as a legal technicality. I want you to close the first-sale doctrine loophole for AI training purposes, require disclosure when AI companies acquire books in bulk, and protect rare and out-of-print works from destruction. Authors, libraries, and future generations have a stake in this. Please act before more of the written record is fed into a machine and thrown away.