AI Company Bulk Purchases Rare and Out-of-Print Books, Scans and Destroys Originals Using High-Speed Machines
AI Company Bulk Purchases and Destroys Rare Books for Training
AI companies are bulk purchasing rare and out-of-print books, scanning them after cutting the spines with high-speed machines, and destroying the originals. Services like ISBNdb facilitate anonymous orders that can reach up to one million copies.
Books published before 2022 are priced higher due to the absence of AI-generated text; a federal judge ruled that the practice of destroying originals while retaining a digital copy constitutes fair use. Anthropic previously hired the former head of Google Books partnerships to advance the "acquire all the books in the world" initiative.
Physical books are irreversibly destroyed and converted into training data, with funds and procurement flowing to intermediaries capable of large-scale scanning and anonymous supply, putting pressure on the existence of rare physical documents.
Source: Public Information
ABAB AI Insight
Companies like Anthropic have received court confirmation in the Bartz v. Anthropic case: Legally purchased physical books that are destructively scanned and have their originals destroyed for training constitute fair use; this ruling reduces the legal risks of obtaining "clean" pre-training text, directly leading to large-scale physical book procurement and destruction pipelines.
In terms of capital flow, internet text has been polluted by AI, making physical books from before 2022 high-value, unpolluted corpora; intermediaries connect with laboratories using anonymity, NDAs, and "digital preservation" rhetoric, shifting resources from traditional book circulation to one-time digital destruction, forming a new supply chain.
A similar path can be seen in the early Google Books scanning controversy: large-scale digitization once sparked copyright and preservation disputes, but at that time, many originals were retained. Current training demands have turned "destruction to prove a single copy" into a legal and commercial incentive.
Essentially, this is a technological substitution: when models require vast amounts of unpolluted text, the uniqueness of physical books is transformed into digital training assets, fundamentally reversing the priorities of preservation and use.
ABAB News · Cognitive Law
Once fair use of destruction is recognized, rare physical copies will accelerate their disappearance
The premium for unpolluted corpora will ultimately be priced by destruction costs
Anonymity and NDAs are standard accompaniments to large-scale irreversible actions