AI Companies Are Buying Tons of Old Books Because They're Free of AI Slop

2026-07-21 · Source: 404media Feed · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Fundamental Awareness, medium

Summary

AI companies are increasingly purchasing vast quantities of old, printed books, primarily those published before 2022, to use as training data for their large language models. This trend is driven by the need for "clean" data, free from AI-generated text that can lead to "model collapse." ISBNdb, a company specializing in book metadata, now facilitates bulk acquisitions of 1,000 to 1 million books for AI labs, emphasizing that print books offer "curated, peer-reviewed, domain-specific human knowledge." This practice often involves "destructive book scanning," where books are destroyed, leading to "optics problems" and secrecy measures by ISBNdb. Despite copyright concerns, a federal judge ruled Anthropic's destructive scanning for internal digital copies as "transformative" fair use. Booksellers report an unprecedented surge in sales of diverse, often rare, and foreign-language books, indicating significant demand from AI entities.

Key takeaway

For Directors of AI/ML seeking high-quality training data, you should prioritize sourcing pre-2022 print materials to mitigate "model collapse" risks from AI-generated content. Be aware that bulk acquisition often involves destructive scanning, which, while legally deemed fair use for internal copies, carries significant "optics problems." You must weigh the benefits of clean data against potential public perception issues and consider implementing robust NDAs for your sourcing operations.

Key insights

Pre-2022 print books are a critical source of "clean" AI training data, free from model-degrading AI-generated content.

Principles

• Data quality is paramount for AI model stability.
• Destructive copying for internal use can qualify as fair use.
• Secrecy is valued to manage public perception.

Method

AI companies acquire bulk print books via services like ISBNdb, often using destructive scanning for efficient digitization, then process them into training datasets.

In practice

• Source pre-2022 print materials for uncontaminated datasets.
• Implement NDAs for sensitive data acquisition strategies.
• Consider destructive scanning for cost-effective digitization.

Topics

• AI Training Data
• Model Collapse
• Destructive Scanning
• Copyright Law
• Fair Use
• ISBNdb

Best for: CTO, VP of Engineering/Data, AI Architect, Tech Journalist, Legal Professional, Director of AI/ML

Related on AIssential

• Companies are purchasing physical books because books published before the widespread adoption of generative AI offer something... — AI's reliance on human-generated data is creating a scarcity crisis, leading to unsustainable practices like destroying old books.
• Hacking of Suno matters because it appears to expose something that generative-AI companies have usually kept hidden: the operational machinery used to assemble a commercial training corpus. — A generative AI company's training data acquisition methods are now a critical legal battleground, not just fair use.
• AI is Eating Its Own Tail And Biting The Hand That Feeds It — AI training on its own outputs creates a feedback loop leading to model collapse and degraded quality.
• Generative AI floods and dilutes the market for books — Generative AI reshapes creative markets by volume, diluting returns for human-authored works, not by quality.
• AI vendors are destroying rare books to feed the chatbot — AI vendors are systematically destroying physical books for training data, utilizing legal rulings and specialized intermediaries.
• The $200 Billion Data Debt: Every Major AI Lab Is Running Out of Fuel and Nobody Is Talking About… — The AI industry faces a critical data debt crisis as high-quality human-generated data is exhausted and synthetic data causes model collapse.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by 404media Feed.