Amazon, which started off selling books, is destroying rare texts to train AI
18 August 2026 · filed under 966ff312c3d3
Amazon is destroying rare books in order to train artificial intelligence models, according to a report carried by TechCrunch AI on August 17. The report was also discussed on Hacker News.
The underlying dynamic, per the summary supplied with the report, is that large language models have already been trained on most material readily available online. Rare and undigitized texts represent material outside that pool. This is offered as the reason such books hold particular value to companies building AI systems.
Amazon’s role in this practice is notable given the company’s origin. It began as an online bookseller before expanding into cloud computing, retail, and artificial intelligence. The report frames the destruction of rare texts as a turn against that history, a company once built on selling books now consuming them to build machine learning systems.
The available material does not detail which titles or collections are affected. It does not specify how the destruction occurs or what internal process within Amazon is responsible. It also does not include any statement from Amazon addressing the report.
The Specola notes the account as reported, without independent verification beyond the cited coverage, and will update the log if further detail becomes available from primary sources.