AI Book Burning? Companies Are Destroying Millions of Books to Feed Chatbots
AI developers are moving to secure large datasets by acquiring physical books in bulk. These books are stripped apart and scanned into digital formats to train new models. This process often leaves the physical copies destroyed and discarded. Intermediaries facilitate these anonymous purchases to meet the demand for high-quality human text that predates widespread AI-generated content.
The used-book market is feeling the effects of this industrial-scale sourcing. Sellers report a sharp uptick in volume, moving from a few dozen books weekly to hundreds. While this provides a profit avenue for old inventory, many express concern that rare or out-of-print titles are being permanently lost once they are scanned and pulped.
Legal battles surrounding these practices remain active. Recent court rulings have characterized the scanning of books for AI training as transformative fair use, though this interpretation faces challenges in other jurisdictions. One major settlement recently required Anthropic to compensate authors for the use of pirated material in training their models.
Prominent industry figures have addressed the growing backlash, with some advocating for the preservation of physical copies rather than destructive scanning methods. As developers continue to prioritize the volume of data in their models, the conflict between digital expansion and the preservation of physical records continues to draw scrutiny.