In a much-anticipated decision, U.S. Third Circuit Judge Stephanos Bibas, sitting by designation on the Delaware District Court, ruled that Ross’s use of Thomson’s materials to train a competing AI search tool constituted copyright infringement. Thomson Reuters Enter. Ctr. GmbH v. Ross Intel. Inc., No. 1:20-cv-613-SB (D.Del. Feb. 11, 2025). This decision reversed in large part the court’s prior 2023 decision denying Thomson’s motions for summary judgment on copyright infringement and a fair use defense.
In its earlier ruling, the court concluded that whether the individual headnotes (summarizing key points of law and case holdings) and Key Number System (numerical framework for classifying content) were original enough to qualify for copyright protection was a question for the jury. The court now took that issue from the jury finding that both the headnotes and Key Number System met the minimum “spark” of originality. It also found actual copying of 2243 headnotes. It left issues relating to Ross’s use of the Key Number System for the jury given that there remained factual disputes.
Ross asserted defenses of innocent infringement, copyright misuse, merger, scenes á faire, and fair use. The court now found that Ross’s defenses did not “hold[] water.” It found innocent infringement did not apply because the infringed works bore a copyright notice. Copyright misuse did not also apply because Ross did not show that Thomson Reuters misused its copyrights to stifle competition. The merger defense was rejected given that “there are many ways to express points of law from judicial opinions.” Scenes á faire was deemed to not fit because it was not necessary to slim down a judicial opinion to Thomson’s headnotes or categorized by key numbers.
As to fair use, the court found that the two most important of the four fair use factors in 17 U.S.C. §107 weighed in favor of Thomson Reuters. First, it found that Ross’s use was commercial and not transformative (e.g., Ross’s purpose was to “create a legal research tool to compete with Westlaw”)(Factor 1). In this regard, the court found that Ross’s AI tool is unlike a generative AI system that generates new content. As the court noted, “[r]ather, when a user enters a legal question, Ross spits back relevant judicial opinions that have already been written” – a process the court felt resembled the way Westlaw’s headnotes and Key Number System were used. It also found that the effect on the market, which the court considered the most important fair use factor, weighed decidedly in Thomson’s favor because Ross intended to develop a market substitute and there was no possible public benefit to save Ross (Factor 4). While it found that the headnotes “are far from the most creative works” (Factor 2) and Ross did not make Westlaw’s headnotes accessible to the public (Factor 3), it found that weighing all the factors together warranted summary judgment for Thomson on fair use.
This ruling while important is narrow in scope. As noted by the court in finding Ross’s use non-transformative, “only non-generative AI is before me today.” In addition, the court found that Ross intended to use its AI tool to compete with Thomson. The court also found that Ross’s copying (which was of numerical data relating to the headnotes) though intermediate was distinguishable from prior intermediate copying cases because the copying was “[not] reasonably necessary to achieve the user’s purpose.” As such, generative AI system or using data to train non-competing systems may warrant a different result.