Skip to content
Submit incident
AII-1000-46489
AllegedThe record is a lawsuit, complaint or accusation that has not been decided.

AI companies buy and destroy rare books for model training

AI companies reportedly purchased rare and used books in bulk through mystery buyer accounts at scale during 2026, then scanned and destroyed them at warehouses to train language models. Booksellers raised concerns about irreplaceable titles being destroyed and the legality and cultural impact of using copyrighted works without permission.

22 August 2026 Β· OpenAI Β· Intellectual property and appropriation
22 August 2026
AII-1000-464892 sources ↓
Quick facts
Category
Intellectual property and appropriation
Country
United States
Source
The Wall Street Journal
Date
22 August 2026
Drafted by
Raidu Incident Detection

What happened

Starting in spring 2026, rare-book dealers in the US and Britain noticed a sudden surge in bulk orders for used and rare books through online marketplaces including Alibris, Biblio and Amazon.com.

The buyers used unusual project names including Red Sparrow Project, Blue Finch Project and Green Parrot Project, ordered dozens to hundreds of books at a time on unrelated subjects, routed them through logistics warehouses, and did not haggle over prices.

Investigation by journalist Emanuel Maiberg revealed that ordered books were being delivered to Amazon warehouses. Books were reportedly scanned and destroyed at industrial scale at facilities associated with AI companies.

Reported impact

Booksellers reported concern that irreplaceable and rare titles, including single-edition books that had not sold for years, were being destroyed for AI training data.

Allan Stypeck of Second Story Books expressed unease about whether the bulk purchases were pulling from circulation the last copies of scarce titles, and whether the books were serving purposes contradictory to the longevity of bookselling businesses.

Index analysis

AI companies have used printed books to train large language models after exhausting digitally available data. According to researchers, AI training on repetitive existing data causes 'model collapse', driving demand for new content sources.

Books are unique sources covering distinct times, places, cultures and audiences across centuries and languages, raising questions about the appropriateness and legality of their use without permission for commercial AI training.

Classification

Category
Intellectual property and appropriation
Industry
IT infrastructure and hosting
Country
United States
Harm status
Actual harm
Intent
Unknown
Timing
Unknown
Harm types
Economic or property
Harmed
Business, Civil society
Failure modes
Training data misuse, Unauthorised scraping
AI systems
ChatGPT, Claude, Gemini

Sources and evidence

This record was researched and written by the Index, based on the following original reporting.

The Wall Street Journal
AI's Need for Content Has Put Rare-Book Dealers in a Bind
22 August 2026
Oman Observer
AI boom sparks rush to buy and scan old books
23 August 2026
Oct 2026Carlsen Verlag sues OpenAI for copyright infringement of "Das NEINhorn"AllegedIntellectual property and appropriationSpiegel OnlineRP OnlineBΓΆrsenblatt+5Oct 2026Argentine media group Perfil sues OpenAI and Microsoft over unauthorized content useAllegedIntellectual property and appropriationPerfilClarinOct 2026Sony and Warner Sue Anthropic Over Unauthorized Training on Copyrighted MusicAllegedIntellectual property and appropriationΠ£ΠšΠ Π†ΠΠ€ΠžΠ ΠœSep 2026Biologist alleges Claude AI used unpublished research in enzyme discoveryAllegedIntellectual property and appropriationAnadolu AjansΔ±CumhuriyetHaberler+1Sep 2026Mathematician Accuses OpenAI of Using Unpublished Research in Model TrainingAllegedIntellectual property and appropriationThe VergeCybernewsQuartz+2