The rise of generative AI has unlocked new frontiers in how machines read, write, and reason, but what happens when this intelligence is built on a foundation of pirated books? In June 2025, a U.S. federal court issued a landmark ruling in Bartz v. Anthropic, a case where bestselling authors sued AI firm Anthropic for copying their copyrighted works without permission to train its large language model, Claude. While the court ruled that using books to train AI can amount to “fair use” under American law, it drew a sharp line when it came to the use of pirated material. This case sets a powerful precedent and opens a different kind of conversation for countries like Pakistan, where copyright enforcement remains weak and fair use exceptions are narrowly defined. How prepared are we to protect our authors, our data, and our narratives in a world where AI is learning from everything, including what it was never allowed to touch? This has indeed become a million-dollar question.

The Case in Focus: Bartz v. Anthropic

In Bartz v. Anthropic, three U.S.-based authors, Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, sued the AI company Anthropic PBC in the Northern District of California for copyright infringement. The claim? That Anthropic had illegally copied their books, not once, but in several stages, to train its generative AI tool, Claude. (Bartz v. Anthropic PBC, No. 3:24-cv-05417-WHA, at 2–3 (N.D. Cal. June 23, 2025))

The court found that Anthropic had downloaded millions of pirated books from illegal online libraries like Books3, LibGen, and Pirate Library Mirror, and later also purchased physical books, scanned them, and added them to a central digital library (id.. at 3–4). From this internal repository, Anthropic selected certain books (including the plaintiffs’) to train successive versions of Claude.

In a detailed ruling, the court observed that using books to train an AI was “transformative” and counted as fair use under U.S. copyright law (id.. at 10). This means that there was no copyright violation on the part of Anthropic for using the plaintiffs’ texts in training their AI, especially because the output never replicated the original works. However, the court also held that retaining pirated books in a permanent digital library was not fair use, even if those copies weren’t used directly in training. That part of the copying remained unlawful (id.. at 12–13).

In short, the court tried to strike a fair balance by allowing AI training to be used fairly because the technology was being used in a new........

© Courting The Law