Technology Law Alerts

June 26th, 2025

‘Spectacularly Transformative’ — and Still Liable:  The AI Copyright Showdown Begins

In the first federal court ruling on whether training generative AI models with copyrighted materials constitutes fair use, U.S. District Judge William Alsup issued a mixed but monumental decision on June 24, 2025, in Bartz et al. v. Anthropic PBC (N.D. Cal., No. 24-05417 WHA). The judge hailed Anthropic’s Claude model as “among the most transformative many of us will see in our lifetimes,” noting its ability to mimic human reasoning and writing by processing millions of digitized texts. He found that the use of copyrighted books to train such models was “spectacularly transformative,” stating that the LLMs “trained upon works not to race ahead and replicate or supplant them — but to turn a hard corner and create something different.”

Yet Judge Alsup drew a firm line when it came to Anthropic’s methods of acquiring training data. “There is no carveout from the Copyright Act for AI companies,” he wrote. Specifically, he condemned the company’s downloading of over seven million books from pirate sites like Books3 and LibGen in 2021 and 2022, describing it as “inherently, irredeemably infringing.” The judge added, “That Anthropic later bought a copy of a book it earlier stole off the internet will not absolve it of liability for the theft, but it may affect the extent of statutory damages.”

Key Rulings

Training AI Is Fair Use:

  • The court found that using copyrighted books to train Claude was transformative under the first fair use factor. Alsup wrote that “[e]veryone reads texts, too, then writes new texts. They may need to pay for getting their hands on a text in the first instance. But to make anyone pay specifically for the use of a book each time they read it, each time they recall it from memory, each time they later draw upon it when writing new things in new ways would be unthinkable,” rejecting the idea that training an LLM to mimic style or quality constituted infringement. He emphasized that Claude outputs did not copy the plaintiffs’ texts verbatim, nor were the authors alleging direct output-based infringement.

Digitizing Purchased Books Is Also Fair Use:

  • Anthropic’s conversion of millions of purchased print books into digital format—by scanning and destroying the originals—was also protected. Judge Alsup reasoned that the format shift “eased storage and enabled searchability,” and that no new or distributed copies were created. He noted this kind of transformation was long accepted under precedent, likening it to microfilming for archival or research purposes.

But Pirating Books? Not So Fast:

  • The court was unequivocal in rejecting Anthropic’s reliance on pirated books for its central research library, even if they were eventually used for transformative purposes. “Pirating copies to build a research library without paying for it… was its own use — and not a transformative one,” Alsup ruled.

What This Means

The ruling is a major legal win for AI developers who train models on lawfully acquired data. Judge Alsup’s clear endorsement of LLM training as “spectacularly transformative” provides legal cover for companies using purchased, licensed, or otherwise legitimately sourced materials. It signals that fair use can protect core AI development processes—so long as developers steer clear of piracy. 

At the same time, the decision potentially opens the door to statutory damages against Anthropic of up to $150,000 per pirated work, depending on findings at trial.  A pending motion for class certification could significantly expand the scale of damages if granted.

Other high-profile copyright cases involving AI, including lawsuits against Meta and OpenAI, are awaiting similar rulings. Many expect the legal questions to reach the Supreme Court in the coming years, as the boundaries of fair use in the age of generative AI continue to evolve.

If you have questions about the court ruling or about AI or other interactive entertainment issues, please contact Sean F. Kane at (212) 705-4845 or skane@fkks.com, S. Gregory Boyd at (212) 826-5581 or gboyd@fkks.com, or any other member of the Frankfurt Kurnit Artificial Intelligence or Interactive Entertainment Groups.