An 'Unfair Use' Argument: AI-Written Books Are Pushing Human Authors Out of the Market

Artificial intelligence companies have long argued that training their models on copyrighted books causes no financial harm to creators. Now there is concrete evidence that refutes that claim.

The dispute began over Anthropic's practice of digitizing books — a federal judge found the process itself to fall under "fair use" (the copyright-law principle of permitted use). But the central question lies elsewhere: what happens after a model is trained and is then used to generate new content?

The U.S. Copyright Office examined exactly this question last year and identified the weakest point in the AI companies' argument: one of the "fair use" criteria tests whether a new work harms the market for the original. The office concluded that using copyrighted works to create commercial content that competes with them in the marketplace could be precisely that kind of harm. All that remained was to prove it — and now the proof exists.

Researchers analyzed more than 14,000 e-books sold on Amazon between 2023 and 2026 and found many titles with a significant share of AI-written text — so many that they narrowed the entire market for other authors. The number of such books rose 19-fold, and the total money spent on them grew 9-fold. In other words, market capacity grew faster than sales volume, and revenue per average book fell, while books with no detected AI text lost market share.

This is exactly the kind of market harm the Copyright Office said courts should watch for. When AI can produce thousands of books at almost no cost, human authors are forced to compete against a flood of content built on their own original work. According to Ed Newton-Rex, head of the nonprofit Fairly Trained, the study "completely kills" the AI companies' argument that they "do no harm to creators."