Destructive scanning of books broadens access to information for the general public.
Are AI companies purchasing books, digitizing them for training data, and wiping the physical copies from circulation? They are, but the situation is not as bleak as some mainstream coverage would have you believe.
Recently, the Demand Progress Education Fund sent a coalition letter to the Federal Trade Commission (FTC) urging the agency to examine whether bulk digitization of books— including rare, out-of-print titles with few surviving copies— constitutes anticompetitive conduct that quietly shields AI firms from competition. The letter centers on Project Panama, Anthropic’s self-described “effort to destructively scan all the books in the world.”
A destructive scan, defined as destroying a physical book in the course of digitization, may sound ominous, yet it represents a highly efficient means of converting printed works into digital form. It also aids in ensuring adherence to copyright law. As federal Judge William Alsup concluded in Bartz v. Anthropic (2025), destructive scanning can qualify as fair use because the one-to-one transformation from one medium to another— without increasing or distributing copies— is more transformative than many fair-use scenarios where the number of copies rises.
The coalition attempts to sidestep this point by noting that a fair-use finding is not a license to foreclose competition. While that is accurate, it does little to bolster their claim that AI companies are violating antitrust rules.
The coalition contends that “pre-AI human-authored text is a finite resource with an inelastic supply,” yet in numerous cases publishers are willing and able to publish additional copies. In other words, supply is elastic. For genuinely out-of-print works (where supply is inelastic), the existing stock is often plentiful. For instance, although the Encyclopaedia Britannica, 15th edition, has not been printed for over a decade, a complete 32-volume set can be found and bought online for a few hundred dollars.
When dealing with materials for which only a single copy remains— i.e., low, inelastic supply— there is no solid basis to believe these inputs are so crucial to AI model development that digitization would create an insurmountable barrier for established players. The field of AI development remains open, as abundant data sources exist beyond scanned books. Projects like Project Gutenberg host and freely distribute more than 75,000 digitized books. Furthermore, training data for AI models also comes from synthetic data, publicly available data, licensed data, and user-generated data, among other origins.
The coalition’s theory of harm fails to meet the legal standard for predatory overbuying as described in the Supreme Court’s decision in Weyerhaeuser Co. v. Ross-Simmons Hardwood Lumber Co. (2007). That ruling addresses the phenomenon of driving up input prices through monopsony power. Present evidence does not show that prices in the market for out-of-print books have risen to a level that only incumbents can afford.
Moreover, bulk digitization yields substantial pro-competitive benefits, including improvements to AI models that benefit consumers, many of which are available at no cost. If copyright law forces an AI company to inadvertently destroy the last copy of a decades-old volume, the knowledge it contains becomes far more useful and widely accessible when integrated into an AI model rather than remaining locked in a single physical copy in one person’s possession.
Booksellers themselves stand to gain: one seller told 404Media that AI firms’ bulk book purchases help him financially by clearing out inventory that would otherwise struggle to move. He also disclosed that the bulk purchases consisted of books with ISBNs—a system introduced in 1970—suggesting that antiquarian works are not the target. This aligns with the reality that such rare volumes, when not housed in a library or museum, are typically acquired at auctions rather than through bulk orders. Moreover, there is little incentive for an AI company to train on an original Shakespeare folio, which costs millions, when the text is freely accessible for use.
Destructive digitization of a single copy from the rich body of out-of-print works does not threaten humanity’s cultural heritage. Additionally, the FTC should not squander its limited resources on pursuing a case that signals vigorous, dynamic competition among AI developers rather than market foreclosure.
