Copyright Ruling and Vendors Incentivize Destructive Book Scanning

ISBNdb is advertising the ability to source up to one million physical books for AI developers, including titles the company describes as non-digitized, rare, and out of print. The firm’s materials allow buyers to filter by ISBN, subject, year, language and edition and recommend a nondisclosure agreement for each engagement.

A June 23, 2025 federal court order reached two findings related to book scanning and model training. On the record before it, the court found that using purchased copies to train certain large language models was transformative fair use. Separately, the order held that converting lawfully purchased print books into non-distributed digital library copies could be fair use when the PDFs replaced the purchased volumes without increasing the library’s total copy count. The opinion treated copies obtained from a pirated central repository differently, leaving those claims for trial.

Court records document a separate, earlier procurement chain tied to a company that in 2024 hired a former head of a major book-scanning project, bought millions of print books, and used service providers to remove bindings, scan pages and discard paper originals. Public records do not link ISBNdb’s commercial offers to that procurement chain, nor do they identify buyers or confirm destructive scanning for specific ISBNdb orders.

ISBNdb’s promotional material describes destructive scanning followed by verifiable destruction or recycling of the originals and argues that pre-2022 print books are less represented in online training corpora. The company acknowledges reputational risk from headlines about destroyed books in its compliance guidance. No public transaction records or disposal verifications tied to named titles have been released.

Labels such as rare or out of print appear in vendor lists but do not alone establish bibliographic scarcity or uniqueness of a particular physical copy. Conservation concerns about bindings, annotations, printings or provenance are distinct from copyright analysis, which focuses on whether copyrighted text was copied and how it was used. Public documentation has not identified named rare or last-surviving copies destroyed in these programs.

A widely circulated online post combined details from the 2025 order, the documented 2024 procurement activity, and ISBNdb’s marketing into a single account that asserted routine rare-book shredding and a price premium for print material used by AI firms. Transaction data and title-level evidence to support those assertions have not been made public.

Taken together, public filings, court records and vendor marketing show a legal and commercial environment in which converting purchased books into digital training data is presented as practical. At present, no public records have established that unique cultural objects have been destroyed as a result of the ISBNdb offers.