Yesterday, the Third Circuit Court of Appeals unsealed its ruling in the Thomson Reuters v. Ross Intelligence case.
In the opinion, the court affirmed the district court’s ruling that Ross Intelligence’s use of Thomson Reuters’s work to train its AI system was not fair use.
Though the headlines are rightly calling the decision a “landmark AI copyright win,” it’s also easy to overstate the importance of this case when it comes to AI and copyright.
Simply put, the outcome of this case, like most fair‑use cases, is heavily fact‑dependent. It is unclear if those facts will carry over to other cases, most notably the New York Times one that is currently pending.
While it is indeed a major win for Thomson Reuters, it is far too early for creators to celebrate. Simply put, it’s entirely possible that this could be the only victory of its kind.
Background of the Case
Thomson Reuters is the owner of the legal research platform Westlaw. Widely used by law firms and law students to find relevant court decisions, the database features not just millions of opinions, but also human‑written “headnotes” for those opinions, which are summaries of the legal issues in the particular case.
Ross Intelligence, which is now defunct, was an AI‑oriented company that aimed to create a competitor to Westlaw. To do this, they were not only going to use the actual opinions but they needed human‑written summaries of these opinions to train against. Initially, they approached Thomson Reuters to license the use of their content, but the company refused.
Instead, ROSS partnered with a company named LegalEASE to create some 25,000 “bulk memos” that would be used to train the model. Those lawyers were told to access the Westlaw database but not to copy and paste content from it. However, according to Thomson Reuters, 2,243 of their headnotes were copied regardless.
In December 2020, Thomson Reuters sued Ross in a Delaware federal court for copyright infringement. Ross shuttered shortly after the lawsuit was filed but the case continued. In September 2023, the judge ruled on dueling motions for summary judgment, saying that the fair‑use question was a matter for the jury. This set the stage for a possible trial.
However, in February 2025, the judge reversed course and ruled in favor of Thomson Reuters, finding that Ross Intelligence’s use of the Westlaw headnotes was not a fair use as a matter of law. This prompted Ross to appeal the decision and that brings us to the Third Circuit Court of Appeals, which affirmed the district court’s ruling.
Initially, the Third Circuit Court of Appeals decision was sealed. However, a day later, the court published the full document explaining its reasoning. That, in turn, is why I (and others) believe that this case may have only a limited impact on the future of AI and copyright.
The Reasoning Behind the Decision
Though the court ruled on several issues, most of the focus, understandably, has been on the fair use analysis that the court conducted.
When determining whether a use is a fair use, the court looks at four factors:
- the purpose and character of your use
- the nature of the copyrighted work
- the amount and substantiality of the portion taken, and
- the effect of the use upon the potential market.
Of those factors, the first and fourth are generally considered to be the most important. However, the court found that the first, third, and fourth factors all favored Thomson Reuters in this case.
On the first factor, the court found that Ross’ use of the Westlaw headnotes was only minimally transformative and was commercial in nature. After all, Ross’ goal was to create its own legal research platform, meaning that the eventual result would be similar to Westlaw’s headnotes.
The second factor slightly favored Ross since the notes had been previously published and were mostly factual in nature. The third factor favored Thomson Reuters since Ross copied the entirety of the headnotes. The fact that Ross only copied a small percentage of the total number of headnotes was not relevant.
Finally, the fourth factor strongly favored Thomson Reuters since Ross was, in effect, creating a competitor to Westlaw. As such, the court found that it could significantly harm the market for Thomson Reuters’ work.
In the end, with the three factors, including the two most important ones, favoring Thomson Reuters, the court found that Ross’ use of the Westlaw headnotes was not a fair use.
Why It May Not Matter
If one thing should stand out in this case, it’s the dates. Ross’ actions were from the late 2010s, before generative AI was widely available. That’s because Ross Intelligence was not a generative AI company. It was using nascent AI tools to generate a competitor to create a legal research platform. Comparing Ross to AI companies like OpenAI and Anthropic is difficult.
This is something that the ruling says directly. In comparing this case to the Bartz ruling, in which a judge found that Anthropic’s use of copyright-protected content was protected under the fair use doctrine. Since Anthropic’s models aim to create “original expression,” the court found that its use of the content was fair use.
Ross, on the other hand, was using it to create “substitutive competition.” As such, the court reached the opposite conclusion.
In short, using AI to create a direct competitor of the content you are training on is less likely to be a fair use than training a general-purpose generative AI model. In short, unless one can successfully argue that a general-purpose AI is a true competitor to the content it is trained on, it’s unlikely that OpenAI or Anthropic have much to worry about from part of this ruling.
That said, there is one interesting wrinkle. The court does acknowledge the emerging market of licensing content for AI training, saying that training on content without permission may usurp that market.
It’s unclear if and how this could be applied to generative AI companies, but it does highlight a potential path forward for creators concerned about how AI is impacting their work.
Bottom Line
This case is a major win for Thomson Reuters and an interesting one in the AI landscape. However, this ruling, at multiple points, separates itself from the rest of the AI lawsuits. Simply put, it’s just not likely to have much of an impact on the broader AI landscape.
Still, there are some interesting takeaways, and I don’t think it’s a ruling that can be outright ignored. It just needs to be contextualized and noted that Ross Intelligence was not and is not a generative AI company. What it was doing was very different, and it shows in the ruling.
That said, I know many creators will welcome legal pushback against AI companies. To that end, there is some reason to celebrate.
However, I wouldn’t get too excited. Even if it were more directly applicable, it would still be just one ruling, in one country.
The legal landscape around AI is going to take decades to settle. Any ruling will be just one of many.
Keep that in mind when reading any coverage about AI and copyright.
Want to Reuse or Republish this Content?
If you want to feature this article in your site, classroom or elsewhere, just let us know! We usually grant permission within 24 hours.