ISBNdb Pitched Pre-LLM Books for AI Training, Then Pivoted Away
On July 21, 404 Media reported that ISBNdb was offering bulk printed-book sourcing for AI training, including orders of up to 1 million titles and material promoted as predating widespread generative AI. By July 28, the service page had been removed; PC Gamer reported that ISBNdb said the offer was part of exploring demand and that the company had pivoted away from that direction.
ISBNdb promoted a bulk printed-book sourcing service for AI developers before removing the offer and saying it had pivoted away. The episode shows how demand for older, traceable text was becoming a commercial data-sourcing pitch, but it does not establish that ISBNdb completed any orders or that its clients destroyed books.
What ISBNdb offered
An archived June 27 capture of ISBNdb's service page advertised orders of up to 1 million titles. Buyers could specify ISBNs, subjects, publication years, languages, editions or authors, and ISBNdb said it would deliver labeled books to a customer's scanning or digitization pipeline. The page targeted AI research labs, model developers and data aggregators.
The service page described acquisition and delivery, not a named customer list or completed transactions. It also did not say that ISBNdb itself would scan or destroy the books. Those boundaries matter because reporting about wider book-buying activity and Anthropic's documented historical scanning program does not prove how an unidentified ISBNdb customer handled a particular shipment.
The offer was removed
404 Media reported on July 21 that ISBNdb was pitching printed books as training material that avoided contamination from AI-generated web text. The company's marketing emphasized older and non-digitized works as dense, edited sources with clearer physical provenance. Publication date is still only a coarse filter: teams would need to verify authorship, editions, rights and data quality rather than assume every older title is suitable.
By July 28, the service page had disappeared. PC Gamer reported that ISBNdb said the page had been part of exploring demand and that the company had chosen to pivot away from that direction. That update makes the offer a recent, short-lived marketing initiative rather than a service that can safely be described as currently available.
What the Anthropic ruling did and did not decide
ISBNdb's archived page invoked Anthropic's book-scanning litigation as support for the sourcing model. In a June 23, 2025 order, U.S. District Judge William Alsup found fair use on the specific record for Anthropic's conversion of lawfully purchased print books into digital library copies when each print copy was discarded and the replacement was not redistributed. The order separately found that pirated copies used to build Anthropic's central library were not justified by fair use.
The ruling was not a blanket approval for every scanning or training program. Rights, acquisition method, copying, retention, distribution and jurisdiction can change the legal analysis. ISBNdb's marketing claims therefore should not substitute for project-specific legal review.
Why this matters for ML teams
The durable lesson is about provenance controls. A physical purchase can produce a receipt and edition record, but a defensible corpus still needs chain-of-custody documentation, rarity and preservation checks, OCR validation, duplication controls, permitted-use records and a way to remove or correct material later.
The withdrawn offer is evidence that physical book sourcing was being productized for AI workflows. It is not evidence that ISBNdb supplied a particular AI company, that every sourced book was scanned destructively or that older print automatically solves copyright and quality risks.
Key Points
- 1An archived ISBNdb page advertised bulk printed-book sourcing for AI workflows, with orders of up to 1 million titles and delivery to scanning pipelines.
- 2By July 28, the page had been removed and ISBNdb was reported to have said it had explored demand and pivoted away from the offer.
- 3The Anthropic ruling was fact-specific; physical provenance still requires rights, preservation, OCR, retention and correction controls.
Scoring Rationale
The short-lived offer is relevant to training-data provenance, copyright planning and preservation controls, but ISBNdb later said it had pivoted away and the retrieved evidence does not identify completed orders or clients.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
