Multiple California Courts Enter Summary Judgment That Using Copyrighted Material to Train AI Platforms is Fair Use

Within a roughly one-week period in late June 2025, two federal judges in the Northern District of California entered summary judgment rulings on the issue of “fair use” in connection with generative AI platforms’ use of copyrighted material.  These are the first two such rulings in the more than three dozen copyright infringement lawsuits pending in U.S. courts.  (An earlier ruling, involving Thomson Reuters and AI startup ROSS, also weighed in on the fair use question, but that case does not involve generative AI.  The US Court of Appeals for the Third Circuit has agreed to hear an interlocutory appeal in the Thomson Reuters case.)

The headlines have largely cast both rulings as wins for the AI platforms, and to a great extent that is accurate; the decisions resolved key unsettled questions in favor of the AI platforms.  But at the same time, both rulings make clear that there are a number of fact-specific, case-by-case issues that preclude—for now—drawing any blanket conclusions about the state of play between copyrighted content and generative AI platforms.

The Fair Use Factors

Under U.S. copyright law, it can be a defense to copyright infringement if the alleged copying constituted a “fair use.”  Section 107 of the Copyright Act specifies four factors to weigh in determining whether there has been fair use:

1. The purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;
2. The nature of the copyrighted work;
3. The amount and substantiality of the portion used in relation to the copyrighted work as a whole; and
4. The effect of the use upon the potential market for or value of the copyrighted work.

The Bartz Ruling

In Bartz v. Anthropic, U.S. District Judge Alsup issued an order on June 23, 2025, granting partial summary judgment on fair use. The ruling encompasses four distinct scenarios as permutations of two variables:

• Pirated copies of books downloaded from repositories known as Books3, LibGen, and PiLiMi, and copies of books where Anthropic purchased and scanned hard copies.
• Use to train LLMs and use to maintain a reference library

The court found that all of the training uses are fair use, as is the use of purchased and scanned books for the library use.  However, the court found that the library use of pirated books was not fair use.

Concerning training use, the court’s analysis of the four fair use factors is as follows:

1. The purpose and character of the use: Finding that use of copyrighted works to train LLMs is “quintessentially transformative,” analogizing to a human reading books and then using that accumulated knowledge to write new works.  The court did not address the distinction between commercial and research uses.  Additionally, for the pirated books, even though the court later noted that “piracy of otherwise available copies is inherently, irredeemably infringing even if the pirated copies are immediately used for the transformative use and immediately discarded,” it did not exclude the pirated works from its fair use finding on factor 1, which it found to weigh in favor of fair use.
2. The nature of the copyrighted work: The court found that all the plaintiffs’ works contained expressive elements and that this factor weighs against a fair use finding.
3. The amount and substantiality of the portion used: Citing the Google Books ruling, the court framed the analysis as focused not on the amount of the work ingested for training, but on the portion that was thereafter made accessible to the public.  And since the plaintiff authors in this case did not allege that there were any AI outputs that copied their books, this factor was found to support fair use.
4. The effect of the use upon the market value of the copyrighted work: The court found that because (as noted) the LLM is not alleged to have output infringing knockoffs or copies of the authors’ work, and further reasoning that copyright law does “not [] protect authors against competition,” this factor also favored fair use.

Concerning the library use and the purchased copies, the court found it important that only a single copy of each book was retained, reasoning that this was a mere format change that was sufficiently transformative.  This reasoning supported fair use for factors 1, 3 and 4. Factor 2 weighed against fair use for the same reasons as the training use.  The court specifically excluded from this ruling any additional copies made from books in the central library that were not used for training; the court indicated the record is too poorly developed to address this issue on summary judgment.

Finally, the court found no fair use for the pirated copies that were made for library use, concluding, “[w]e will have a trial on the pirated copies used to create Anthropic’s central library and the resulting damages, actual or statutory (including for willfulness).” 

The Kadrey Ruling

On June 25, 2025, Kadrey v. Meta Platforms, Inc., Judge Chhabria, also of the Northern District of California, entered partial summary judgment in favor of Meta on the issue of fair use.

Judge Chhabria framed the issue this way: “Because the performance of a generative AI model depends on the amount and quality of data it absorbs as part of its training, companies have been unable to resist the temptation to feed copyright-protected materials into their models—without getting permission from the copyright holders or paying them for the right to use their works for this purpose.  This case presents the question whether such conduct is illegal.  Although the devil is in the details, in most cases the answer will likely be yes.”

Reacting directly to Judge Alsup’s analysis of the transformative nature of using copyrighted works to train LLMs being analogous to human learning, Judge Chhabria disagreed with Judge Alsup’s analysis, as applied to potential market effects, stating, “This inapt analogy is not a basis for blowing off the most important factor in the fair use analysis.”  Focusing particularly on the market impact factor, the court emphasized the limited nature of its ruling, stating, “this ruling … stands only for the proposition that these plaintiffs made the wrong arguments and failed to develop a record in support of the right one.”

The court went on to analyze the fair use factors:

1. The purpose and character of the use: Again, the court found that because the use to train LLMs is transformative, this factor favors Meta.  But the court also noted that this factor could vary in weight depending on particular facts (for example, if a special-purpose LLM is “designed to be used to create works substantially similar to those on which it was trained.”  The court also addressed the commercial-versus-research aspect of factor one, as well as whether copies of the books were downloaded from “shadow libraries.”  Ultimately, the court found factor 1 to favor fair use.
2. The nature of the copyrighted work: The court found that this factor weighs against a fair use finding, but also noted that it is entitled to relatively less weight.
3. The amount and substantiality of the portion used: The court adopted a different analysis from the Bartz ruling, finding that the factor favors fair use, “even though [Meta] copied the plaintiffs’ books in their entirety.  The amount that Meta copied was reasonable given its relationship to Meta’s transformative purpose.”
4. The effect of the use upon the market value of the copyrighted work: Calling this the “single most important element of fair use,” the court outlined a number of different ways that copying for AI training could harm the market for the copyrighted works, concluding that even generation of “works that are similar enough (in subject matter or genre) that they will compete with the originals and thereby indirectly substitute for them” is a promising theory of harm.  However, the court found that plaintiffs’ presentation on this issue was too weak to stave off summary judgment.

Conclusions

Taken together, these two rulings highlight some key points of agreement and of disagreement.  They present a roadmap of potential additional points of argument in pending and future cases, particularly focused on the balance of facts impacting factor 1, and especially inviting parties to more fully develop an analysis of the market impacts for factor 4.

These are just two rulings in the forty-odd pending cases, with more cases likely.  The application of the law to this stunning and disruptive new technology has plenty of runway to continue to develop.

[View source.]