Mid-Year Review: AI Copyright Case Developments in 2025
The first half of 2025 saw major developments in some of the dozens of ongoing copyright lawsuits against AI companies, as well as the filing of a few new high-profile cases. Two cases in particular, Bartz v. Anthropic and Kadrey v. Meta, saw orders on summary judgment that sent shockwaves through the copyright community when courts in the Northern District of California found that certain types of generative AI training qualified as fair use. However, as we’ve written, the orders are complex, conflicting at times, and likely have a limited impact on other cases. This blog looks back on those orders and other key developments in AI and copyright cases over the past six months.
Background
There have been over 50 AI infringement lawsuits filed over the past few years, although the current number of active cases is somewhere closer to 30 due to many consolidations and a few settlements. About half of the ongoing cases are proposed class actions, and almost all of those are by groups of authors of literary works against large language model developers. There are also cases brought by visual artists, news publishers, YouTube video copyright owners, and computer programmers. The types of works involved in a majority of the cases are literary works, but there are also cases involving music compositions and song lyrics, sound recordings, computer code, photographs, videos, and databases. Most of the cases have either been brought in the Northern District of California or the Southern District of New York, although there are also cases in the District of Delaware and the District of Massachusetts. A list of all the different cases and their status can be found here.
The lawsuits are all in different stages, procedurally, and some of the earlier filed cases have had to amend their complaints and narrow down the claims—as some of the more tenuous claims related to AI output and derivative works have been rejected. Importantly, claims related to direct infringement at the ingestion stage have largely survived, and we’re now beginning to see some of the first decisions on fair use.
Two Major Court Decisions on Fair Use Raise More Questions than Answers
Bartz v. Anthropic
On June 24, a court in the Northern District of California issued an order on summary judgment in an ongoing case brought by authors of literary works against AI company Anthropic for the unauthorized use of their works for AI training. The order assessed fair use as applied to three separate acts: (1) the use of the works to train a generative model, (2) the conversion of purchased print copies to digital, and (3) the downloading of pirated copies of books to build a “library.” The court found that the first two qualified as fair use while the third did not, and it concluded that there will be a trial on the pirated copies and the resulting damages.
Determining that generative AI training is “exceedingly” transformative and thus qualifies as fair use, the court glossed over the foremost fourth fair use factor, misinterpreted the Supreme Court’s instruction in Warhol v. Goldsmith, and made numerous other legal errors in its analysis (which are discussed in detail in this blog post). Reporting on the decision, many commentors chalked it up as a big win for generative AI companies, but, because Anthropic could be on the hook for damages in the billions, it’s difficult to label this decision as a win of any type. While the fair use analysis related to training is troubling for copyright owners, the reasoning is flawed on multiple levels and likely to be corrected on appeal. What should be very concerning to AI companies is the court’s conclusion that Anthropic’s copying of pirated copies of works only some of which were subsequently used for training is not fair use. If that same determination is applied widely across similar generative AI infringement cases, AI companies could face staggering damages.
Anthropic has since filed a motion for an interlocutory appeal on the piracy issue and attempted to stay the district court proceedings until the Ninth Circuit hears their appeal. The court recently denied a stay of the case, which is set to go to trial in December. Moreover, the court has certified the class of plaintiffs, meaning that Anthropic could face crippling statutory damages. Ultimately, the outcome of the case could have a significant impact on generative AI infringement litigation, where many AI developer defendants have ingested training material from the same piracy-laden datasets.
Kadrey v. Meta
Just two days after the Bartz order, on June 25, another court in the Northern District of California issued an order on summary judgment in a case against Meta, finding that the use of the plaintiffs’ books for training Meta’s large language model (LLM) was “highly transformative” and qualified as fair use. While the holding was detrimental to the plaintiffs in the case, the court made abundantly clear that its analysis was very narrow due to it being based on a lack of evidence presented by counsel. The decision itself may prove to be very helpful to copyright owners as it basically provided a roadmap for how they can win a copyright infringement case against a generative AI company. In fact, in many ways the case represents a victory for all copyright owners other than the 13 plaintiffs in the case because of the court’s clear signaling that it would have ruled differently had certain issues been raised and evidence presented.
What may prove to be the most important aspect of the decision was the court’s very thoughtful and lengthy discussion of the indirect substitutional impacts that could harm the copyright owners’ actual and potential markets. Unlike Bartz, the Kadrey court seems to understand and appreciate that the advent of generative AI technology and the scale of its impact on copyright’s incentives to create and distribute new copyrighted works for the public to enjoy requires a more thoughtful approach that is more consistent with the spirit and purpose of copyright and the fair use defense.
Surprisingly, unlike Bartz, the court in Kadrey excused Meta’s use of pirated works from shadow libraries because it concluded that (i) there was no evidence that Meta’s acts benefitted “these libraries or perpetuated their unlawful activities—for instance, if they got ad revenue from Meta’s visits to their websites;” and (ii) use of these pirated works was justified because they were eventually used by Meta for AI training purposes which he found to be a transformative use. This approach has no basis in the law and is tremendously misguided and dangerous because, among other things, it would essentially immunize a defendant for engaging in massive acts of piracy so long that they can show that somewhere downstream the pirated copies were used for a purpose that qualifies as fair use.
While the summary judgment order was a setback for the plaintiffs, the case is proceeding on the issue of whether Meta simultaneously uploaded copyrighted works while using BitTorrent technology to download them—a process known as seeding. If the court finds that Meta distributed massive amounts of pirated copyrighted works, it could be held liable for massive damages, similar to Anthropic.
New AI Copyright Cases and the Multidistrict Litigation
The steady stream of AI infringement lawsuit filings continued in the first half of 2025, with a mix of similar complaints against LLM developers and a new, high-profile action against the image generator Midjourney.
Advance Local Media, et al. v. Cohere, Inc.
On February 13, a group of news and magazine publishers—including Advance, The Atlantic, Guardian, Los Angeles Times, Politico, Forbes, and Vox Media—filed a complaint against AI company, Cohere, for the unlicensed copying, use, and dissemination of the publisher plaintiffs’ news and magazine articles to train Cohere’s suite of large language model AI systems called the “Command Family” of models. According to the complaint, Cohere’s AI models deliver outputs that include full verbatim copies, substantial excerpts, and substitutive summaries of the publishers’ works. The publishers also point out that the retrieval-augmented generation (RAG) feature in Command routinely returns verbatim copies of publishers’ copyright protected works in response to user queries. The court is currently considering Cohere’s motion to dismiss.
Disney Enterprises Inc. v. Midjourney Inc.
The case that seems to be making the most headlines recently was brought on June 11 by Disney and Universal when they filed a complaint against Midjourney in the Central District of California for direct and secondary copyright infringement related to the company’s AI image generating service. Calling Midjourney “the quintessential copyright free-rider and a bottomless pit of plagiarism,” the complaint alleges that Midjourney ignored the plaintiffs’ demands to stop using their works or, at a minimum, adopt technical measures that would prevent the generation of infringing material. The complaint further alleges that Midjourney reproduces, publicly displays, and distributes copies and derivatives of characters from popular Marvel and Star Wars franchises, among others, and it provides visual evidence of dozens of instances of infringing output. The case is one of only a few that involve claims against an image generator for infringing output—the other being Getty v. Stability AI, which has been moving slowly through discovery—and it will be one to watch given the very strong evidence presented and the fair use defense Midjourney has raised. Midjourney has since filed an answer to the complaint, arguing among other things, that its use of plaintiffs’ works for training qualifies as transformative fair use.
Bird v. Microsoft Corporation
On June 24, a group of authors, including Kai Bird, Jia Tolentino, Daniel Okrent, and others filed a complaint against Microsoft in the Southern District of New York over the use of pirated copies of plaintiffs’ books to train the AI model, Megatron. The complaint alleges that Microsoft used a dataset of nearly 200k pirated books that were copied to train Megatron to enable the AI model “to generate a wide range of expression that mimics the syntax, voice, and themes of the copyrighted works on which it was trained.” The case is likely to be centralized with many others in the Southern District of New York against OpenAI and Microsoft as part of the multidistrict litigation (MDL) discussed below.
Strike 3 Holdings, LLC et al v. Meta Platforms, Inc.
On July 23, adult content company Strike 3 Holdings filed a complaint against Meta in the Northern District of California over the torrenting and unlicensed use of 2,396 of plaintiffs’ adult movies to train Meta’s AI models. Strike 3 stated that through its anti-piracy tools, it traced the IP addresses that infringed its adult movies back to Meta-owned IP addresses and that a two year-analysis indicated that off-site servers were being used for Meta’s AI activities to torrent and infringe plaintiffs’ works. Given the allegations of Meta’s use of BitTorrent technology to download, and possibly simultaneously upload, pirated copyrighted works, the case could spell big trouble Meta.
Multidistrict Litigation (MDL)
In April, the U.S. Judicial Panel on Multidistrict Litigation (MDL) centralized over a dozen cases across the country in the District Court for the Southern District of New York for pretrial activity, which could include dispositive motions including summary judgment, as well as discovery disputes that have been common among the cases. The case, which is now known as In re OpenAI Copyright Infringement Litigation, includes high profiled cases brought by Raw Story Media, the Intercept, the Authors Guild, and the New York Times, many of which were already consolidated. The MDL also now includes two other case filed in the first half of 2025, Denial v. OpenAI and Ziff Davis v. OpenAI, which involve infringement claims brought by a group of authors of literary works and a group of online publishers, respectively. Because the cases were centralized and not consolidated, they are still subject to orders and other actions that are specific to a particular case. However, there is now one judge, Judge Sidney Stein, overseeing the litigation and there may be broad orders that affect the various plaintiffs.
Thomson Reuters v. Ross Intelligence
One final case worth mentioning is a long-running copyright infringement case brought by Thomson-Reuters (TR) against Ross Intelligence, a competitor legal research service, for copyright infringement related to Ross’s scraping of copyrighted legal content from TR’s Westlaw platform to develop its own competing product based on machine learning. The case doesn’t involve the same type of generative AI technology at the heart of many of the large language model or image generator lawsuits, but it has raised important questions about copyrightability and fair use as applied to training.
Originally scheduled to go to trial last year, the judge postponed trial at the last minute and allowed the parties to bring new summary judgment motions on copyrightability, validity, infringement and fair use. In early 2025, the judge granted TR’s motion for summary judgment on direct copyright infringement and related defenses against Ross. The court found that TR’s Westlaw headnotes are copyrightable and rejected Ross’ fair use defense, explaining that the use was commercial and not transformative (under factor 1) and that Ross’ use harmed the potential market for AI training data (under factor 4). However, in April, the court granted Ross’s motion for interlocutory appeal to the Third Circuit on the issues of copyrightability and fair use.
Despite not being a “generative” AI case, the appeal has garnered a lot of attention from the copyright community because the Third Circuit’s fair use approach could still influence how other courts analyze the unauthorized use of copyrighted works for training. To that end, a number of amicus briefs are expected in support of both sides in the coming months.
Looking Ahead
The second half of 2025 could see more major developments in the various ongoing AI infringement lawsuits, although the next decisions on fair use are not likely to come until 2026. The Bartz v. Anthropic case is set to go to trial in late 2025, and while a decision isn’t likely until 2026, all eyes will be on the trial and parallel interlocutory appeal. Attention will also be focused on developments in the Kadrey v. Meta case over whether the plaintiffs can show that Meta distributed millions of copies of pirated works, potentially exposing it to billions in damages. It will also be interesting to see what happens in the MDL and whether the Southern District of New York splits from the approach to generative AI training taken by the Northern District of California. Stay tuned for what’s sure to be an action-packed end to 2025.
If you aren’t already a member of the Copyright Alliance, you can join today by completing our Individual Creator Members membership form! Members gain access to monthly newsletters, educational webinars, and so much more — all for free!