AI vendors are destroying rare books to feed the chatbot

2026-07-23 · Source: Pivot to AI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Novice, long

Summary

AI vendors are actively destroying rare physical books to acquire training data for chatbots, a practice exemplified by Anthropic's "Project Panama." In 2025, Judge William Alsup ruled Anthropic's destructive scanning of purchased books as "exceedingly transformative" and thus fair use, despite the company's efforts to keep the initiative secret. However, Anthropic faced a \$1.5 billion payout for training on pirated book libraries, which was not deemed fair use. The company reportedly continues to acquire rare books for destructive scanning. Bulk book buyers like Zoom Books are suspected of facilitating this process, purchasing large volumes of old books. ISBNdb, a book catalog platform, is now a key part of this chain, marketing pre-2022 print books as a "provably clean corpus" for AI training, advocating for their scanning and pulping, and assuring AI clients of anonymity. ISBNdb justifies this by stating a physical book's purpose is served once its information is encoded into an AI model.

Key takeaway

For AI Ethicists and Legal Professionals evaluating data sourcing, you must acknowledge the Anthropic ruling. This precedent permits destructive scanning of physical books as fair use for AI training. Services like ISBNdb now facilitate bulk acquisition and pulping of pre-2022 texts. This practice demands critical examination of intellectual property and cultural preservation. It also raises ethical questions about treating physical artifacts as mere data sources. Advocate for policies balancing AI innovation with safeguarding historical knowledge.

Key insights

AI vendors are systematically destroying physical books for training data, utilizing legal rulings and specialized intermediaries.

Principles

• Destructive scanning for AI training can be fair use.
• Pre-2022 print books offer "clean" training data.
• Anonymity is crucial for AI data sourcing.

Method

AI companies employ bulk buyers and catalog platforms to acquire, scan, and pulp physical books, integrating extracted text into models.

In practice

• Target pre-2022 print materials for clean data.
• Engage specialized book catalog services.
• Understand "transformative use" in data acquisition.

Topics

• AI Training Data
• Fair Use Doctrine
• Copyright Law
• Book Preservation
• Large Language Models
• Anthropic
• ISBNdb

Best for: Investor, CTO, VP of Engineering/Data, AI Ethicist, Legal Professional, Tech Journalist

Related on AIssential

• AI Companies Are Buying Tons of Old Books Because They're Free of AI Slop — Pre-2022 print books are a critical source of "clean" AI training data, free from model-degrading AI-generated content.
• Gracenote Media Services, LLC v. OpenAI: a database/metadata case aimed at the infrastructure layer of AI quality—identifiers, taxonomies, curated relational logic, and editorial descriptions. — Metadata copyright suits target AI's "truthy" backbone, challenging free ingestion of curated B2B data.
• The Danger of “Just Scrape It” in AI Strategy — Illegitimate data acquisition, even for transformative AI use, creates severe legal and architectural risks.
• The Guardian view on changes to copyright laws: authors should be protected over big tech | Editorial — Authors are demanding robust copyright legislation to protect their work and livelihoods from unlicensed AI use.
• BMG sues Anthropic for using Bruno Mars, Rolling Stones lyrics in AI training — AI training on copyrighted content faces increasing legal challenges, particularly from aggressive copyright holders like music labels.
• Who’s Afraid of Chinese Models? — A new U.S. copyright policy could declare AI training data collection fair use and prohibit model distillation bans.

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 Pivot to AI.