Independent booksellers are currently seeing a strange surge in large, bulk orders for secondhand novels. Many shop owners, including Stuart Manley of Barter Books, report shipping thousands of items to international warehouses with no clear destination. These orders vary widely, ranging from obscure Latin academic texts to mass-market cowboy fiction. The lack of specific demand for these titles suggests the purchasing is driven by something other than traditional reader interest.
Evidence points toward the growing appetite of artificial intelligence firms for training data. A 2025 court ruling in the United States determined that using purchased books to train AI models does not violate copyright law, provided the usage is transformative. Internal documents from the firm Anthropic revealed a project known as Project Panama, which aims to scan books at an industrial scale. This process often involves destructive scanning, where the spines of books are removed to facilitate rapid digitization, followed by the recycling of the remaining pages.
This trend has created a complex situation for the book trade. On one hand, sellers are moving inventory that has sat idle on shelves for decades. Booksellers acknowledge that the industry has limited need for millions of copies of common titles. However, concerns remain regarding the potential loss of rare or historically significant editions that might be caught up in these bulk acquisitions.
Legal experts note that regulations regarding data ingestion for training models vary significantly by region. While US law currently permits this practice under specific judicial interpretations, copyright laws in the United Kingdom and other jurisdictions maintain stricter requirements for obtaining permission from rights holders before reproducing protected works. As technology companies continue to prioritize diverse datasets to refine large language models, the secondhand book market finds itself in an unexpected position at the center of the global AI development debate.