AI Firms Fuel Secondhand Book Boom via Copyright Loophole

The secondhand book market is experiencing an unexpected surge as AI developers turn to physical shelves to feed their data-hungry models. This trend follows a landmark 2025 US court ruling that fundamentally altered the legal landscape for training Large Language Models (LLMs).

In a decision delivered by Judge William Alsup, the court established that the destructive scanning of legally purchased secondhand books for AI training qualifies as fair use. The ruling specifies that this practice does not violate US copyright law, provided the resulting digital files are maintained internally and not distributed to the public. Consequently, AI firms—most notably Anthropic—have spent millions of dollars acquiring physical books in bulk to digitize them. This strategy allows companies to avoid the costly, time-consuming process of negotiating licensing agreements or the legal risks associated with digital piracy.

The First Sale Loophole

This shift is driven by the "first sale doctrine," a legal principle that allows the owner of a lawfully acquired physical copy of a work to sell, lend, or destroy that copy without the permission of the copyright holder. While traditionally used to support used bookstores and libraries, the doctrine is now being leveraged by tech giants to create massive new datasets. By purchasing the physical object, AI companies bypass the traditional licensing fees that authors and publishers typically receive when their work is used digitally.

Industry Implications

For the AI industry, this approach provides a reliable stream of high-quality text data at a time when digital archives are becoming more guarded and litigated. However, the trend has created significant tension within the literary community. While some booksellers benefit from the sudden spike in demand, others express deep ethical concerns. The practice of destructive scanning—where books are destroyed during the digitization process—poses a risk to cultural heritage, particularly if rare volumes are bought specifically for destruction after surviving for generations.

What to Watch

As AI firms continue to scale their data acquisition, the focus now shifts to the long-term impact on the physical book market. Booksellers report a wide and seemingly random variety of titles being purchased, suggesting a broad appetite for diverse linguistic data. The industry is now watching to see if further legal challenges will emerge to narrow the scope of the first sale doctrine or if this physical-to-digital pipeline will become the standard for the next generation of AI training.