More than a dozen U.S. civil-society organizations are urging the Federal Trade Commission (FTC) to investigate major AI companies over reports that they are buying large quantities of physical books, scanning them for AI training and then destroying the originals. The groups argue that this should be examined not merely as a copyright or cultural-preservation issue, but as a potential antitrust problem.
The central argument is that destroying the physical books could amount to a form of “input foreclosure.” High-quality human-written books are a valuable training resource, particularly older works that predate widespread generative AI. If wealthy AI companies acquire large quantities of these scarce physical sources and permanently remove them from circulation, competitors—including smaller AI startups—could face higher costs or reduced access to comparable training material.
The groups specifically point to reports about Anthropic's Project Panama, under which the company acquired millions of books, removed their bindings, scanned them and discarded the physical copies. They argue that the practice could contribute to a broader concentration of AI capabilities by allowing the largest companies to secure valuable training inputs while preventing those resources from remaining available to competitors.
This creates an unusual intersection between AI training, intellectual property, competition policy and preservation of knowledge. Copyright disputes generally ask whether companies have the right to use particular works for training. The antitrust argument asks a different question: even if a company can legally acquire a book, can it deliberately remove a scarce resource from the market in a way that makes competition harder? The civil-society groups want the FTC to investigate whether the reported practices constitute an unfair method of competition under Section 5 of the FTC Act.
The broader significance is that the AI competition debate is expanding beyond chips, computing power and cloud infrastructure to include the raw materials needed to build models. If high-quality human knowledge becomes a strategic input, controlling access to that knowledge could become another source of market power.
