Judge Rules 'Destructive Scanning' of Books Legal for AI Training

A federal judge has ruled that "destructive scanning"—the process of purchasing physical books, removing their bindings for digitization, and discarding the remains—is legal under fair use standards. The decision arrives as AI companies face intensifying scrutiny over how they acquire the massive datasets required to train large language models (LLMs).

The ruling in Bartz v. Anthropic determined that the physical destruction of legally acquired copyrighted works for the purpose of data extraction does not violate copyright law. This legal victory for AI developers stands in stark contrast to other copyright disputes. Specifically, Anthropic recently reached a $1.5 billion settlement with a group of authors to resolve claims regarding the company's use of pirated e-books to train its AI models.

The Data Hunger of LLMs

Large language models require vast amounts of text to evolve and maintain accuracy. While much of the internet is readily available for scraping, a significant portion of human knowledge remains locked in physical print. To bridge this gap, some AI companies have turned to physical libraries and bookstores, employing high-speed scanning processes that often render the original physical copy unusable.

Anthropic has stated that its AI service, Claude, is trained on a combination of publicly available web data and commercially acquired datasets, alongside self-generated data. This shift toward synthetic data is a growing trend in the industry. Alexis Ramsey-Tobienne, the Assistant Dean for AI and Learning Integrity at Eckerd College, notes that AI is increasingly "learning from itself," being trained on files that include outputs the AI has previously created.

A New Precedent for Fair Use

This ruling establishes a critical legal boundary for the AI era. By distinguishing between the legal purchase of physical media and the acquisition of pirated digital files, the court has created a roadmap for how AI companies can legally ingest physical archives. The decision suggests that once a company owns a physical copy of a work, the act of destroying that copy to extract its data for a transformative purpose—such as training a neural network—falls under the protection of fair use.

However, the ruling highlights a growing tension between technological acceleration and the preservation of cultural artifacts. While the law may permit the destruction of a purchased book, the practice remains controversial among archivists and bibliophiles who view the physical book as more than just a data carrier.

Future Implications

Industry observers are now watching to see if this precedent will encourage other AI firms to aggressively purchase and digitize physical archives. While the Bartz v. Anthropic ruling provides a shield for destructive scanning, the $1.5 billion settlement serves as a warning that the use of unauthorized digital repositories remains a high-risk legal gamble. The industry now awaits further litigation to determine where the line between "transformative use" and copyright infringement truly lies.