How AI is coming for second-hand books

Second-hand book sales are experiencing an unexpected surge, with booksellers in multiple countries reporting a sharp increase in bulk orders. These orders, often highly specific and targeting niche categories such as New Zealand cookbooks, tourist pamphlets, and historical texts, have raised concerns among sellers. In particular, bookstores in Wales, England, Ireland, India, the United States, Europe, Australia, and New Zealand have noticed an uptick in unusual requests. One notable example comes from Wellington’s Book Haven, which began receiving large orders from a Canadian company earlier this month. Owner Anne-Marie Thorby investigated and found that the same buyer was purchasing tens of thousands of books from around the globe, all destined for a single American warehouse. This activity has sparked speculation about the true intent behind these purchases. The connection to artificial intelligence emerged following revelations about Project Panama, an initiative by AI firm Anthropic aimed at training its language model, Claude. First exposed by The Washington Post in January 2026, Project Panama involves using a hydraulic-powered cutting machine to remove spines from physical books, allowing their pages to be scanned and integrated into AI training datasets. Internal documents obtained by the newspaper described the project as an effort to “destructively scan all the books in the world” and emphasized that the company wished to keep the operation confidential. Thorby, having become aware of Project Panama, linked the recent bulk orders to the initiative. She expressed concern that these actions could undermine the cultural and historical value of books, stating, “Our role is to keep books alive, to keep them in the world.” Her comments reflect growing unease among independent booksellers who see themselves as custodians of literary heritage rather than mere suppliers of raw materials for corporate AI projects. The scale of the operation remains unclear, though estimates suggest the number of books involved could reach into the hundreds of thousands. A vendor’s project proposal cited by The Washington Post indicated that Anthropic was seeking a document scanning service capable of processing between 500,000 and two million books over a six-month period. This highlights the potential magnitude of the undertaking, raising questions about the environmental and ethical implications of such mass-scale book destruction. In response to the controversy, an Anthropic spokesperson clarified that the company acquires books from mainstream commercial markets and that training language models on textual content is a common practice. However, the spokesperson added that “none of our data acquisition programs buy and destroy rare or antiquarian books.” This statement contrasts with reports from booksellers who claim to have observed the procurement of rare and specialized works. The situation also intersects with ongoing legal battles involving AI firms and intellectual property rights. Authors and publishers have filed lawsuits against companies including Anthropic, Meta, Google, and OpenAI, alleging that these entities are systematically harvesting copyrighted material without consent. Some experts argue that older books, free from AI-generated content, are particularly valuable for training high-quality language models. Additionally, U.S. copyright laws may allow AI companies to legally acquire and use second-hand books for training purposes without explicit permission from rights holders. As the debate continues, booksellers and advocates warn that the increasing demand for physical books in the context of AI development could lead to unintended consequences. While the exact scope and future direction of these activities remain uncertain, the growing involvement of major tech firms in the book trade underscores a broader shift in how literature is valued and utilized in the digital age.