AI companies’ purchase and scanning of millions of books to train their models has become one of the most contested issues in copyright law following court rulings in the United States. According to Melahat Boran, a lawyer, arbitrator and secretary-general of the Türkiye Press and Publishing Professional Association, US lawsuits indicate that scanning a book may legally constitute “reproduction.”
The Anthropic case and a $1.5 billion settlement
Boran recalled that in the Bartz v. Anthropic case in the United States, training models with books that had been legally purchased and scanned was deemed to fall under “fair use.” However, training with copies downloaded from pirate websites and stored in a centralized digital library was ruled an infringement, and the case ended in a $1.5 billion settlement.
Boran cited “Project Panama” as the best-documented example. Under the project, a contractor removed the bindings from books using hydraulic cutters and scanned the pages; the digital files were delivered to the company, while the printed books were sent for recycling. A supplier’s proposal projected that between 500,000 and 2 million books could be processed within six months.
Proposed solution: collective licensing
Emphasizing that innovation and the rights of authors and publishers are not alternatives to one another, Boran argued that licensing would provide rights holders with predictable income and developers with legal certainty. She cited the voluntary, non-exclusive collective licensing model of the US-based Copyright Clearance Center as an example.
In Europe, the Munich Regional Court found in cases brought by the music copyright organization GEMA that a model’s “memorization” and recognizable reproduction of works in its outputs constituted reproduction. A ruling by the Court of Justice of the European Union in the Like Company-Google case referred from Hungary is awaited to establish binding precedent on the issue.
The situation is different in Türkiye
According to Boran, Türkiye’s Law No. 5846 on Intellectual and Artistic Works recognizes the reproduction of a work by any means as the exclusive right of the author, and has no equivalent of the text and data mining exception in the European Union. Purchasing a book only allows its resale or gifting; it does not authorize reproducing it by scanning.
Boran proposed requiring AI companies to be transparent about their training data, obliging publishers to record unusual bulk orders, and establishing a viable collective licensing system. The World Intellectual Property Organization’s first Global Forum on intellectual property and artificial intelligence will be held in Geneva on November 9-10, 2026.
