The AI Shredder: Trading Human History for Training Data

AI companies are legally purchasing and shredding rare historical books to harvest high-quality training data that predates the era of AI-generated content. Facilitated by anonymous services and protected by 'fair use' rulings, this practice ensures that only a digital version remains while the physical artifact is destroyed. The author warns that this represents an irreversible loss of cultural heritage that cannot be replaced or recovered.

Key Points

• AI companies are systematically shredding rare, pre-2022 books to obtain 'pure' training data free of AI-generated content.
• The service ISBNdb facilitates anonymous bulk orders of up to a million books, using NDAs to keep the practice hidden from public scrutiny.
• A federal legal ruling has deemed the destruction of these books 'fair use' under the logic that it maintains a one-to-one ratio between the original and the digital copy.
• Major AI firms like Anthropic are aggressively pursuing the digitization of all available literature, often hiring former library and database experts to lead the effort.
• Unlike digital scraping, this process is irreversible, leading to the permanent loss of unique historical artifacts and rare botanical or scientific texts.

Sentiment

The community is broadly critical of the reported practice and skeptical that private AI ingestion counts as preservation. However, the reaction is not a simple consensus: many commenters agree with the article's ethical concern while others challenge the evidence, narrow the harm to genuinely rare artifacts, or frame the issue as a broader copyright and archival-policy failure rather than an AI-only scandal.

In Agreement

• Destroying books after scanning them for private AI corpora is cultural vandalism because the public loses access to the original artifact while companies capture the value.
• A model or private training dataset is not preservation, since it may be inaccessible, lossy, uninspectable, and unable to retain provenance or material context.
• The practice exposes broken incentives in copyright and publishing, where works can be kept scarce or out of print instead of being made available through durable public editions.
• Physical books can have future scholarly, historical, or artistic value that is not obvious at the time of destruction, so treating obscure works as disposable is shortsighted.
• If digitization is justified as preservation, the scans should be deposited in public archives rather than locked inside corporate training pipelines.

Opposed

• The article may not prove that genuinely rare or irreplaceable books are being destroyed, and some commenters suspect sensational framing around ordinary destructive scanning.
• For many mass-printed or obscure books, the textual content may be the main value, so digitization can be a net preservation gain even if one physical copy is lost.
• Libraries and institutions already discard books at scale, so blaming AI companies alone may ignore broader failures in archival funding, storage, and access.
• Some commenters argue that owners generally have the legal right to destroy property they bought, even if others find the choice distasteful.
• The outrage may overstate the importance of every old printed object and understate the practical cost of nondestructive scanning and indefinite storage.