Two Northern District of California courts handed down decisions last week in two key copyright lawsuits that challenged the use of copyrighted print and digital books in training the large language models (LLMs) that are at the heart of generative artificial intelligence (Gen-AI) platforms. These two decisions represent the second and third decisions to determine whether copying written materials when training LLMs is “fair use.” Both decisions found at least some copying was “fair use” given the “highly transformative” nature of LLMs and the Gen-AI they support. The decisions leave open future lawsuits, however, against companies who use pirated material to train their LLMs and also allow potential arguments where a plaintiff can show a harmful impact on the market for authors’ works.
On June 23, 2025, Judge William Alsup issued a summary judgment decision in Bartz et al. v. Anthropic PBC, holding the defendant’s LLM was “quintessentially transformative” of the material on which it was trained, fulfilling the central requirement for protection under the fair use doctrine. The case was brought by three plaintiffs, each of whom are authors whose copyrighted works were used by Anthropic in building its “central library” and on which its LLM was trained. Anthropic’s library contained millions of items, including both “copyrighted books in digital form from pirate sites on the internet” and “purchased copyrighted books” that were later digitized.
Although finding copying of material to train an LLM was fair use, Judge Alsup’s opinion found the method by which the defendant obtains the copyrighted material matters. The order distinguished between material that was purchased legitimately versus material downloaded for free from pirate websites. The court allowed the latter claims to continue to trial with the implication that they are infringing copies and a trial is needed to ascertain damages. Because the statutory damages allowed if the jury finds the infringement as willful can exceed one trillion dollars, this case is far from over.
Two days after Judge Alsup’s summary judgment order, Judge Vince Chhabria filed a summary judgment decision that centered on the defendant’s fair use defense. In Kadrey et al. v. Meta Platforms Inc. et al., a group of 13 best-selling authors sued Meta for using copyrighted material without permission to train its LLMs.
Although Judge Chhabria also found the output of the Gen-AI platforms was transformative, he nevertheless seemed ready to rule the defendant’s activity was not protected as fair use because the impact it could have on the market for the plaintiff’s own written works. In short, if Gen-AI can create a written work in the same style as an author in mere minutes and charge a fraction of what the author would charge for a work that took them years to create, the fair use defense may not be available. Because this harm remained hypothetical, however, Judge Chhabria found the fair use defense in this specific case applied.
What this means to you
|
If AI-related copyright litigation were a baseball game, it would barely be out of the first inning and the score is barely 2-1. These two decisions could nevertheless exert significant influence on the development of case law as courts will have to distinguish or adopt the reasoning. Developers of AI platforms should pay close attention to the ebb and flow of the multiple AI-related copyright cases making their way through the legal system, but these initial decisions hold encouraging signals that fair use could be a viable defense when LLMs are trained on purchased copyrighted material.
Contact us
If you have questions regarding the intersection of copyright law and the development of Gen-AI platforms, contact Dustin Taylor or your Husch Blackwell attorney.
|