Reports about AI companies destroying rare books have triggered widespread alarm. While it is confirmed that organizations like Anthropic have previously acquired and destroyed physical books to build training sets, current rumors about a massive, global purge of rare libraries remain unverified. The core motivation for these companies is the acquisition of high-quality, pre-2022 human text. AI models trained on their own synthetic output experience significant performance degradation, so companies are hunting for stable, human-authored data from the 1970s through the 1990s.
Investigations into large bulk orders reveal a complex logistics chain. Booksellers report receiving massive requests from entities like Zoom Books, which often routes inventory through third-party logistics firms such as PrepFort. These companies maintain strict confidentiality agreements, preventing public disclosure of their end clients. While some entities like 2077AI have directly contacted booksellers with specific lists of academic texts, the connection between these purchases and large-scale destruction is not always clear.
The books being acquired are often not rare in a traditional sense. Instead, they consist of out-of-date technical manuals, specialized guides, and academic niche publications. These items hold little value for collectors but offer the dense, factual content that AI models require to avoid the so-called model collapse. Booksellers frequently note that the shipping costs for these transactions exceed the market value of the books themselves.
Regardless of whether these specific volumes are being shredded, the industry shift toward digitizing physical media for machine learning is real. The process of stripping authorship from works and blending them into large-scale models creates a new type of cultural friction. This practice masks the labor of human authors and centralizes control over information, fundamentally altering how society values traditional written knowledge. Whether the books are preserved or pulped, the act of ingesting them into a closed AI system carries its own form of permanence that is currently being debated across the publishing world.