Amazon destroys rare books, according to report, as it builds datasets for AI models

Amazon, once known for being an online bookseller, is reportedly destroying rare books rather than keeping them for resale or reuse, a practice the report links to Amazon’s efforts to train large language models (LLMs) that learn patterns from text.

The report, carried by Yahoo Finance, centers on the idea that LLMs are trained on what is already available online. In that framing, rare physical books would be less useful to model training than the text that can already be captured through other sources, according to the account described in the article.

Amazon has long used machine learning across its business, including recommendations, advertising targeting, and cloud services through AWS (Amazon Web Services). More recently, the company has also increased public attention on generative AI, where models can produce text, summaries, and other outputs based on their training data.

In the marketplace, the reported decision highlights a tension between preserving physical cultural assets and building the large-scale digital datasets that modern AI systems require. Rare books are valuable for collectors and libraries, but the article suggests that for AI training, their practical value is limited if their content is already represented elsewhere online.

The situation also underscores how quickly the economics of media distribution have changed. Amazon’s retail identity began with books and grew into a broader platform for digital content and third-party sales, but the training-data requirements of LLMs add a new layer to how text sources can be valued.

An Amazon newsroom page does not, in the material consulted for this story, provide direct detail on the reported rare-book destruction practice. The company’s public communications in general cover AI, AWS, retail, and workplace topics, but they do not appear to address this specific handling of physical books in the information reviewed here.

Still, even if the operational details are limited, the reported approach fits a broader pattern in AI dataset management: companies must decide which sources to keep, which to digitize, and how to handle copyrighted or scarce materials. LLM training also involves governance questions, including how data is obtained, how long it is retained, and what processes are used to comply with legal and policy obligations.

For now, the public record remains incomplete. The Yahoo Finance report does not, in the information reviewed for this story, provide enough detail to confirm the scope of the practice, the exact operational steps Amazon uses, or whether the affected materials include specific titles, editions, or acquisition channels. What Amazon says about the dataset sourcing and retention rules will be key for assessing the implications beyond the headline.