Recently, a federal court issued the first ruling on the closely watched issue of fair use in copyright infringement involving AI. The court ruled in favor of the plaintiff on its direct infringement claim, and ruled that the defendant’s use of plaintiff’s material to train its AI model was not a fair use.
The Upshot
- On February 11, 2025, the court in Thomson Reuters v. Ross Intelligence reconsidered its prior decision that the question of fair use needed to be decided by the jury and instead ruled on renewed summary judgment motions that defendant’s use was not fair use.
- The case involved defendant’s alleged infringement of Thomson Reuters’ Westlaw headnotes. Ross licensed “Bulk Memos” from a third party to train Ross’s AI-based legal search engine. The “Bulk Memos” were created using Westlaw headnotes.
- The court found that the headnotes were original and copyrightable, and granted summary judgment to Thomson Reuters on direct infringement for certain headnotes.
- On Ross’s fair use defense, the court found that the use was commercial and not transformative. It also found that the use impacted both the legal research market and the market for data to train AI tools. Overall, the fair use analysis favored Thomson Reuters.
- Courts are just starting to reach decisions in AI-based copyright cases. The fair use analysis provides guidance for how future courts will think about these issues.
The Bottom Line
This closely watched decision is significant as it’s the first of its kind so far in the landscape of AI-related copyright litigation. While the infringement finding is fairly specific to the facts of the case, the fair use ruling will likely be influential for future courts’ analysis of this defense, particularly its discussion of the purpose and market impact of using copyrighted materials to train AI models. Ballard Spahr lawyers closely monitor this area of law to advise clients on issues of artificial intelligence and copyright infringement.
On February 11, 2025, Third Circuit Judge Stephanos Bilbas, sitting by designation in the District of Delaware, issued a summary judgment decision in the closely watched copyright infringement dispute between Thomson Reuters and Ross Intelligence concerning Ross’s AI-based legal search engine. The court granted most of Thomson Reuters’ motion on direct copyright infringement, and held that Ross’s defenses, including fair use, failed as a matter of law. This case is significant as it’s the first of its kind to address fair use in connection with artificial intelligence, though the court was careful to point out that this matter, unlike many others working their way through the court system, involved a non-generative AI system.
The underlying case concerns Ross’s AI-based legal search engine and Thomson Reuters’ claim that the use of Thomson Reuters’ Westlaw headnotes as training material for the AI tool constituted copyright infringement. Thomson Reuters’ Westlaw platform contains editorial content and annotations, like headnotes, that guide users to key points of law and case holdings. Ross, a competitor to Westlaw, made a legal research search engine based on artificial intelligence, and initially asked to license Westlaw content to train its product. When Thomson Reuters refused, Ross turned to a third party, LegalEase, which provided training data in the form of “Bulk Memos” consisting of legal questions and answers. The Bulk Memos were created using Westlaw headnotes.
Thomson Reuters brought claims of copyright infringement based on this use. In 2023, the court largely denied Thomson Reuters’ motions for summary judgment on copyright infringement and fair use, and held that those issues were properly decided by a jury. After reflection, the court “realized that [its] prior summary-judgment ruling had not gone far enough,” and invited the parties to renew their summary judgment briefing. This time, the court largely ruled in Thomson Reuters’ favor.
First, the court held that Thomson Reuters’ headnotes were sufficiently original to be copyrightable, even if they were based on the text of underlying court cases. The court found that”[i]dentifying which words matter and chiseling away the surrounding mass expresses the editor’s idea about what the important point of law from the opinion is,” and therefore has enough of a “creative spark” to overcome the low bar presented by the originality requirement. “Similarly, Westlaw’s Key Numbering System was also sufficiently original, as Thomson Reuters had chosen a particular way to organize these legal topics, even if it was not a novel one. The court then turned to actual copying and substantial similarity and granted summary judgment to Thomson Reuters on headnotes which “very closely track[ ] the language of the Bulk Memo question but not the language of the case opinion.” Other headnotes and the Key Numbering System were left for trial.
On fair use, the court granted summary judgment for Thomson Reuters, finding that Ross’s use was not fair. On the first fair use factor, the purpose and character of the use, the court found that Ross’s use was commercial and served the same purpose of Thomson Reuters’: a legal research tool. In the parlance of fair use law, Ross’s use was not “transformative.” The court also rejected Ross’s analogy to earlier computer programming cases where intermediate copying was necessary, and rejected Ross’s argument that the copying was allowed because the text of the headnotes was not reproduced in the final product.
The second and third factors (nature of the material and how much was used), went to Ross, but the fourth factor, the likely effect on the market for the original work, and “the single most important” of the four factors, went to Thomson Reuters. The court looked at both the current market for the original work and potential derivative ones, and found that Ross’s use impacted both the original market for legal research and the derivative market for data to train AI tools. The court found that it did “not matter whether Thomson Reuters has used the data to train its own legal search tools; the effect on a potential market for AI training data is enough.” Altogether, the four fair use factors favored Thomson Reuters, and it was granted summary judgment on fair use.
Looking beyond this opinion, it is the first decision to substantively address fair use in the context of artificial intelligence, so it will be an important guidepost for the multiple cases pending across the country, many of which involve companies who have used copyrighted works to train generative AI models. However, the opinion has an important caveat, which is that “only non-generative AI” was at issue in the case. Generative AI models use their training data set to create new text, image, video, or other outputs. Non-generative models, by contrast, analyze and classify data based on patterns learned from their training data. The cases involving generative AI may involve different analysis for the fair use factors like the question of transformativeness and the nature of the original works, but the opinion’s commentary on current and potential markets, as well as its willingness to weigh the four factors on summary judgment, may be highly applicable.
In short, this is an important decision but much remains unsettled in the law applying copyright to artificial intelligence. Ballard Spahr lawyers closely monitor developments concerning artificial intelligence and intellectual property, including copyright infringement and fair use. Our AI Legislation and Litigation Tracker provides a comprehensive view of AI-related legislative activities and important information about litigation matters with significant potential impact on clients.