As the first copyright cases concerning AI reach appeals courts, EFF wants to protect important, beneficial uses of this technology—including AI for legal research. That’s why we weighed in on the long-running case of Thomson Reuters v. ROSS Intelligence. This case raises at least two important issues: the use of (possibly) copyrighted material to train a machine learning AI system, and public access to legal texts.  

ROSS Intelligence was a legal research startup that built an AI-based tool for locating judges’ written opinions based on natural language queries—a competitor to ubiquitous legal research platforms like Lexis and Thomson Reuters’ Westlaw. To build its tool, ROSS hired another firm to read through thousands of the “West headnotes” that Thomson Reuters adds to the legal decisions it publishes, paraphrasing the individual legal conclusions (what lawyers call “holdings”) that the headnotes identified. ROSS used those paraphrases to train its tool. Importantly, the ROSS tool didn’t output any West headnotes, or even the paraphrases of those headnotes—it simply directed the user to the original judges’ decisions. Still, Thomson sued ROSS for copyright infringement, arguing that using the headnotes without permission was illegal.  

Early decisions in the suit were encouraging. EFF wrote about how the court allowed ROSS to bring an antitrust counterclaim against Thomson Reuters, letting them try to prove that Thomson was abusing monopoly power. And the trial judge initially ruled that ROSS’s use of the West headnotes was fair use under copyright law. 

The case then took turns for the worse. ROSS was unable to prove its antitrust claim. The trial judge issued a new opinion reversing his earlier decision and finding that ROSS’s use was not fair but rather infringed Thomson’s copyrights. And in the meantime, ROSS had gone out of business (though it continues to defend itself in court).  

The court’s new decision on copyright was particularly worrisome. It ruled that West headnotes—a few lines of text copying or summarizing a single legal conclusion from a judge’s written opinion—could be copyrighted, and that using them to train the ROSS tool was not fair use, in part because ROSS was a competitor to Thomson Reuters. And the court rejected ROSS’s attempt to avoid any illegal copying by using a “clean room” procedure often used in software development. The decision also threatens to limit the public’s access to legal texts. 

EFF weighed in with an amicus brief joined by the American Library Association, the Association of Research Libraries, the Internet Archive, Public Knowledge, and Public.Resource.Org. We argued that West headnotes are not copyrightable in the first place, since they simply restate individual points from judges’ opinions with no meaningful creative contributions. And even if copyright does attach to the headnotes, we argued, the source material is entirely factual statements about what the law is, and West’s contribution was minimal, so fair use should have tipped in ROSS’s favor. The trial judge had found that the factual nature of the headnotes favored ROSS, but dismissed this factor as unimportant, effectively writing it out of the law. 

This case is one of the first to touch on copyright and AI, and is likely to influence many of the other cases that are already pending (with more being filed all the time). That’s why we’re trying to help the appeals court get this one right. The law should encourage the creation of AI tools to digest and identify facts for use by researchers, including facts about the law. 

Jimmy Kimmel has been in the news a lot recently, which means the ongoing lawsuit against him by perennial late-night punching bag/convicted fraudster/former congressman George Santos flew under the radar. But what happened in that case is an essential illustration of the limits of both copyright law and the “fine print” terms of service on websites and apps. 

What happened was this: Kimmel and his staff saw that Santos was on Cameo, which allows people to purchase short videos from various public figures with requested language. Usually it’s something like “happy birthday” or “happy retirement.” In the case of Kimmel and his writers, they set out to see if there was anything they couldn’t get Santos to say on Cameo. For this to work, they obviously didn’t disclose that it was Jimmy Kimmel Live! asking for the videos.  

Santos did not like the segment, which aired clips of these videos, called “Will Santos Say It?”.  He sued Kimmel, ABC, and ABC’s parent company, Disney. He alleged both copyright infringement and breach of contract—the contract in this case being Cameo’s terms of service. He lost on all counts, twice: his case was dismissed at the district court level, and then that dismissal was upheld by an appeals court. 

On the copyright claim, Kimmel and Disney argued and won on the grounds of fair use. The court cited precedent that fair use excuses what might be strictly seen as infringement if such a finding would “stifle the very creativity” that copyright is meant to promote. In this case, the use of the videos was part of the ongoing commentary by Jimmy Kimmel Live! around whether there was anything Santos wouldn’t say for money. Santos tried to argue that since this was their purpose from the outset, the use wasn’t transformative. Which... isn’t how it works. Santos’ purpose was, presumably, to fulfill a request sent through the app. The show’s purpose was to collect enough examples of a behavior to show a pattern and comment on it.  

Santos tried to say that their not disclosing what the reason was invalidated the fair use argument because it was “deceptive.” But the court found that the record didn’t show that the deception was designed to replace the market for Santos’s Cameos. It bears repeating: commenting on the quality of a product or the person making it is not legally actionable interference with a business. If someone tells you that a movie, book, or, yes, Cameo isn’t worth anything because of its ubiquity or quality and shows you examples, that’s not a deceptive business practice. In fact, undercover quality checks and reviews are fairly standard practices! Is this a funnier and more entertaining example than a restaurant review? Yes. That doesn’t make it unprotected by fair use.  

It’s nice to have this case as a reminder that, despite everything, the major studios often argue, fair use protects everyone, including them. Don’t hold your breath on them remembering this the next time someone tries to make a YouTube review of a Hollywood movie using clips.  

Another claim from this case that is less obvious but just as important involves the Cameo terms of service. We often see contracts being used to restrict people’s fair use rights. Cameo offers different kinds of videos for purchase. The most well-known comes with a personal use license, the “happy birthdays,” and so on. They also offer a “commercial” use license, presumably if you want to use the videos to generate revenue, like you do with an ad or paid endorsement. However, in this case, the court found that the terms of service are a contract between a customer and Cameo, not between the customer and the video maker. Cameo’s terms of service explicitly lay out when their terms apply to the person selling a video, and they don’t create a situation where Santos can use those terms to sue Jimmy Kimmel Live! According to the court, the terms don’t even imply a shared understanding and contract between the two parties.  

It's so rare to find a situation where the wall of text that most terms of service consist of actually helps protect free expression; it’s a pleasant surprise to see it here.  

In general, we at EFF hate it when these kinds of contracts—you know the ones, where you hit accept after scrolling for ages just so you can use the app—are used to constrain users’ rights. Fair use is supposed to protect us all from overly strict interpretations of copyright law, but abusive terms of service can erode those rights. We’ll keep fighting for those rights and the people who use them, even if the one exercising fair use is Disney.  

Things are speeding up in generative AI legal cases, with two judicial opinions just out on an issue that will shape the future of generative AI: whether training gen-AI models on copyrighted works is fair use. One gets it spot on; the other, not so much, but fortunately in a way that future courts can and should discount.

The core question in both cases was whether using copyrighted works to train Large Language Models (LLMs) used in AI chatbots is a lawful fair use. Under the US Copyright Act, answering that question requires courts to consider:

1. whether the use was transformative;
2. the nature of the works (Are they more creative than factual? Long since published?)
3. how much of the original was used; and
4. the harm to the market for the original work.

In both cases, the judges focused on factors (1) and (4).

The right approach

In Bartz v. Anthropic, three authors sued Anthropic for using their books to train its Claude chatbot. In his order deciding parts of the case, Judge William Alsup confirmed what EFF has said for years: fair use protects the use of copyrighted works for training because, among other things, training gen-AI is “transformative—spectacularly so” and any alleged harm to the market for the original is pure speculation. Just as copying books or images to create search engines is fair, the court held, copying books to create a new, “transformative” LLM and related technologies is also protected:

[U]sing copyrighted works to train LLMs to generate new text was quintessentially transformative. Like any reader aspiring to be a writer, Anthropic’s LLMs trained upon works not to race ahead and replicate or supplant them—but to turn a hard corner and create something different. If this training process reasonably required making copies within the LLM or otherwise, those copies were engaged in a transformative use.

Importantly, Bartz rejected the copyright holders’ attempts to claim that any model capable of generating new written material that might compete with existing works by emulating their “sweeping themes, “substantive points,” or “grammar, composition, and style” was an infringement machine. As the court rightly recognized, building gen-AI models that create new works is beyond “anything that any copyright owner rightly could expect to control.” 

There’s a lot more to like about the Bartz ruling, but just as we were digesting it Kadrey v. Meta Platforms came out. Sadly, this decision bungles the fair use analysis.

A fumble on fair use

Kadrey is another suit by authors against the developer of an AI model, in this case Meta’s ‘Llama’ chatbot. The authors in Kadrey asked the court to rule that fair use did not apply.

Much of the Kadrey ruling by Judge Vince Chhabria is dicta—meaning, the opinion spends many paragraphs on what it thinks could justify ruling in favor of the author plaintiffs, if only they had managed to present different facts (rather than pure speculation). The court then rules in Meta’s favor because the plaintiffs only offered speculation. 

But it makes a number of errors along the way to the right outcome. At the top, the ruling broadly proclaims that training AI without buying a license to use each and every piece of copyrighted training material will be “illegal” in “most cases.” The court asserted that fair use usually won’t apply to AI training uses even though training is a “highly transformative” process, because of hypothetical “market dilution” scenarios where competition from AI-generated works could reduce the value of the books used to train the AI model..

That theory, in turn, depends on three mistaken premises. First, that the most important factor for determining fair use is whether the use might cause market harm. That’s not correct. Since its seminal 1994 opinion in Cambell v Acuff-Rose, the Supreme Court has been very clear that no single factor controls the fair use analysis.

Second, that an AI developer would typically seek to train a model entirely on a certain type of work, and then use that model to generate new works in the exact same genre, which would then compete with the works on which it was trained, such that the market for the original works is harmed. As the Kadrey ruling notes, there was no evidence that Llama was intended to to, or does, anything like that, nor will most LLMs for the exact reasons discussed in Bartz.

Third, as a matter of law, copyright doesn't prevent “market dilution” unless the new works are otherwise infringing. In fact, the whole purpose of copyright is to be an engine for new expression. If that new expression competes with existing works, that’s a feature, not a bug.

Gen-AI is spurring the kind of tech panics we’ve seen before; then, as now, thoughtful fair use opinions helped ensure that copyright law served innovation and creativity. Gen-AI does raise a host of other serious concerns about fair labor practices and misinformation, but copyright wasn’t designed to address those problems. Trying to force copyright law to play those roles only hurts important and legal uses of this technology.

In keeping with that tradition, courts deciding fair use in other AI copyright cases should look to Bartz, not Kadrey.