The past few weeks have seen a flurry of activity in the legislature and the courts relating to Artificial Intelligence (“AI”).
Since the founding of our nation, there has been ongoing debate about whether legal matters should be handled at the federal or state level. The recent budget bill from the House of Representatives, sent to the Senate, included a provision that would have blocked federal funding for AI infrastructure or broadband internet to states that did not impose a 10-year ban on AI regulation. Many governors, attorneys general, and legislators from both sides of the aisle opposed it, stating that, in the absence of a federal AI Act, states must be able to protect their residents, especially children and teens. On June 30th, Republican Senator Marsha Blackburn from Tennessee and Democratic Senator Maria Cantwell from Washington introduced an amendment to remove that provision, and the Senate voted 99 to 1 to do so. Many states already have laws related to AI, and none are willing to relinquish their authority to do so. For now, since there is no federal AI Act, each state retains the right to legislate on this important topic.
One matter before the Courts in the last few years is whether AI companies can use copyrighted content generated by third parties to train their generative AI platforms without paying for it. As expected, copyright owners sued. The owners argue that such use constitutes copyright infringement; the AI companies claim that using the copyrighted content is a fair use. What is the status of these cases? And what is the applicability of each decision nationwide?
Recent Cases Disputing AI Use of Copyrighted Content
The first decision regarding the use of copyrighted content to train AI was issued this past February involving Thomson Reuters Westlaw and an AI start-up. Thomson Reuters Westlaw’s platform uses headnotes (short summaries of points of law in a legal opinion), which are tied to a key number; the key number, in turn, enables the user to access a list of cases that reveal the same or similar legal point. Westlaw owns a registered copyright for its headnotes and key numbers. Ross is an AI startup that was refused a license from Westlaw to train its competing AI platform, but it used the headnotes and key numbers anyway, arguing that its use was transformative and qualified for the fair use defense. Notably, the Ross platform is not generative AI, which creates new content by learning from the training content. Thomson Reuters Enterprise Centre GmbH et al. v. Ross Intelligence Inc., 765 F. Supp.3d 382 (D. Del. 2025)
The question of transformation is the key to the AI training cases given the Supreme Court’s 2022 decision in Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, et. al, 598 U.S. 508 (S. Ct. 2023), which evaluated whether the purpose and character of the use of Goldsmith’s photographs by Warhol as a magazine cover was transformative. The Supreme Court chose not to focus its inquiry on whether the magazine cover itself was transformative in comparison to the original photographs. Accordingly, in Thomson Reuters v. Ross, District Court Judge Stephanos Bibas decided that Ross’s use of the headnotes and key numbers to create a competitive product was not transformative and harmed the market for data to train legal AI platforms, finding for Thomson Reuters in its summary judgment motion for direct copyright infringement. Ross has appealed the decision to the U.S. Court of Appeals for the Third Circuit.
More recently, district courts in the Northern District of California issued two decisions regarding the AI transformation issue, and the waters are indeed murky. In Bartz v. Anthropic, 3:24-cv-05417 (N.D. Cal. 2025), a group of authors filed a class action complaint against Anthropic for direct copyright infringement for Bartz’s use of their copyrighted content to train its Claude large language model (“LLM”). An LLM is a deep learning generative AI system trained on vast datasets to understand and generate comprehensible human language in response to specific questions, known as prompts. Claude is an AI assistant designed to help people code, write, and talk with one another.
Initially, Bartz knowingly used a pirated set of the authors’ books; later, Bartz purchased the authors’ print books, created digital copies, and trained Claude on the content. On June 23, 2025, District Court Judge William Alsop issued an interesting Janus-type decision, first finding that Anthropic’s use of the pirated books constituted direct copyright infringement, which will lead to a jury trial on damages. Second, Judge Alsop found that the use of the purchased books for training was a “spectacular” transformation and a fair use, since people have read books for centuries, and creating an AI platform does not supplant the ability to read books. Judge Alsop also found that Bartz’s creation of digital copies was fair use since it replaced the print books it threw out after digitizing them – a mere format change, not within the scope of exclusive rights granted to authors under the Copyright Act. Interestingly, in Judge Alsop’s view, the author’s market for licensing their works to train generative AI platforms is not one covered by the Copyright Act. Does this finding overlook the reasoning in Warhol, which requires a court to consider the commerciality of the actual use? This decision will likely be appealed.
On June 25, 2025, District Court Judge Vincent Chhabria handed a victory to Meta in Kadrey v. Meta Platforms, Inc., 23-CV-03417-VC (N. D. Cal. 2025), deciding that Meta’s use of pirated books to train its Llama AI models is not a direct copyright infringement. Plaintiff authors include Ta-Nehisi Coates, Junot Diaz, Richard Kadrey, and Sarah Silverman. Contrary to Judge Alsop’s finding that the use of pirated books is not fair, Judge Chhabria said "There is no serious question that Meta’s use of the plaintiffs’ books had a 'further purpose' and 'different character' than the books—that it was highly transformative," Judge Chhabria demolished the plaintiffs’ market harms argument: "Meta has defeated the plaintiffs’ half-hearted argument that its copying causes or threatens significant market harm," said Chhabria. "That conclusion may be in significant tension with reality, but it’s dictated by the choice the plaintiffs made... while failing to present meaningful evidence on the effect of training LLMs like Llama with their books on the market for [AI-generated] books."
Closing Thoughts and Future Considerations
There is hope for copyright owners in other cases, based on other comments by Judge Chhabria:
In cases involving uses like Meta's, it seems like the plaintiffs will often win, at least where those cases have better-developed records on the market effects of the defendant's use. No matter how transformative LLM training may be, it's hard to imagine that it can be fair use to use copyrighted books to develop a tool to make billions or trillions of dollars while enabling the creation of a potentially endless stream of competing works that could significantly harm the market for those books. And some cases might present even stronger arguments against fair use. For instance, as discussed above, it seems that markets for certain types of works (like news articles) might be even more vulnerable to indirect competition from AI outputs. On the other hand, though, tweak some facts and defendants might win. For example, using copyrighted books to train an LLM for nonprofit purposes, like national security or medical research, might be fair use even in the face of some amount of market dilution.
These decisions will likely be appealed, first to the Ninth Circuit Court of Appeals and then to the Supreme Court. For the time being, these decisions apply only in their respective districts and to the specific parties in these cases.
For copyright owners with content, now is the time to ensure you have copyright registrations for your content and to establish a market for AI training, so that you have the tools to negotiate license agreements with AI platforms.
[View source.]
