The high-profile copyright case between Thomson Reuters and Ross Intelligence was thrust back into the spotlight this month when a three-judge panel of the U.S. Court of Appeals for the Third Circuit heard arguments over whether ROSS’s use of Westlaw headnotes to train legal research systems was protected by fair use.
During oral arguments held in Philadelphia on June 11, the three-judge panel, consisting of Judges L. Felipe Restrepo, Tamika R. Montgomery-Reeves, and Emil J. Bove, focused many of its questions on the first and fourth elements of fair use: whether the use of ROSS is transformative and harms the actual or potential market for Thomson Reuters’ copyrighted works.
Earlier this week, the court released the transcript of the hearing.
On behalf of ROSS, White & Case partner Mark S. Davis asked the court to reverse the district court’s decision and find the use of the headnotes to be fair use.
West Publishing, represented by Thomson Reuters and Kirkland & Ellis partner Dale M. Sendari, argued that ROSS copied protected editorial content to build a competing legal research product and that the district court correctly rejected its fair use defense.
The argument comes in an appeal of a Delaware federal court ruling in a case that has garnered attention across the legal tech industry due to its implications for AI training and copyright law. Although the case predates the current generation of generative AI tools, ROSS argued that its technology is an early example of the same broader AI transformation currently reshaping legal research.
ROSS claims fair use
Davis begins by citing a recent decision from the Third Circuit Court of Appeals. ASTM vs. Upcodebased on that precedent, the court argued that it should be immediately revoked on the basis of fair use. He said ROSS built its legal search platform in 2015 by adapting headnotes to questions and using them to create training materials for a system that connects natural language legal questions to judicial opinions.
But Judge Restrepo quickly pivoted the discussion to the fourth fair use element, asking Davis to address the market impact of the use. Mr. Davis responded that the copyrighted works at issue were headnotes themselves, and that there was no market for headnotes as single headnotes or as AI training materials.
Judge Bove asked whether there is or could be a market for AI training. Davis argued that the use of ROSS is too specific to define a market around it. He said ROSS didn’t just take the headnotes as raw data, but instead built training notes containing questions and multiple answers (both correct and incorrect) to train the model.
He argued that having Thomson Reuters claim that it had lost a market for its use was circular because any copyright holder could say it had licensed the very use that was later claimed to be fair.
The judge repeatedly questioned Davis about how ROSS’s product differed from Westlaw’s from a user’s perspective. Judge Bove asked what made ROSS’s system transformative “as opposed to just another type of legal search engine.”
As Davis focused on internal technology, Judge Restrepo noted that ROSS was still a direct competitor to Westlaw, and Judge Montgomery Reeves asked whether the product was intended to serve as a commercial replacement.
Davis acknowledged that ROSS competes with Westlaw, but said copyright law is not designed to protect incumbents from competition. He described ROSS’s product as one in which users ask legal questions in natural language and receive a series of answers in the form of quotes from judicial opinions. Because the output is based on court opinions rather than generated text, there is “no danger of hallucinations,” he said.
Judge Bove continued to press for a factual explanation of the product, at one point telling Davis that there was “a lot of dancing going on” and that he was trying to understand what Ross was trying to do and why it was materially different from Westlaw.
“Could you explain to me what the engine that Ross proposed is, how it’s different, and why it’s so different from what happens when you log into Westlaw in the morning?” Judge Bove asked.
Davis responded that from a user’s perspective, both products help users find cases, but internally ROSS works differently because it uses deep learning technology to teach the system the semantic relationships of legal terms.
“The question involves legal terminology, and what we taught this machine is how to think like a lawyer,” Davis said.
The Committee also questioned Mr. Davis about the copyrightability of the headnotes themselves. Justice Montgomery-Reeves asked why a headnote that condensed and reformulated judicial language was not sufficiently original to qualify for copyright protection.
Davis argued that judicial opinions belong to the public and that the differences between the opinions and the headnotes were too small to support protection. However, the judge examined whether the opinion’s word choice and summary reflected editorial judgment.
Judge Bove highlighted one example in which West’s editors summarized excerpts from an opinion. He suggested that even if headnotes closely track opinion, editors make judgments about what is salient and what should be included. Mr. Davis argued that Mr. West’s editorial instructions were designed to eliminate creativity and track opinion language as precisely as possible.
Davis argued that ROSS should prevail with respect to the second and third fair use elements because the headnotes were true and because ROSS copied only a small portion of Westlaw’s complete headnotes. He said ROSS used only 0.08% of Westlaw’s 28 million head notes.
“We received very little,” he said. “Of the 28 million headnotes, only 0.08% were copied.”
When Judge Montgomery Reeves asked what the outcome would be if Thomson Reuters prevailed in factors one and four and ROSS prevailed in factors two and three, Judge Davis said ROSS should still win. Judge Restrepo asked, “If you take 2 and 3 and your opponent takes 1 and 4, do you win?” Mr. Davis responded that fair use is a multifactor test.
Judge Bove also questioned Mr. Davis about allegations of bad faith by Thomson Reuters, including communications involving Mr. Dentons and efforts by Ross to obtain Westlaw materials. (The record does not detail how Dentons was involved, and no reference to it was found in the parties’ redacted briefs.)
Davis said ROSS paid for the original database and notes, and the licensing issue should be addressed in a separate contract claim, not a fair use analysis. He cited examples that argue that copyright is not reserved only for “well-behaved people” and argued that even evidence of malicious intent should not invalidate fair use.
“Typical substitution example”
Cendari, arguing for Thomson Reuters, began by addressing the scope of the appeal. He said the district court granted summary judgment on 2,834 headnotes, of which Judge Stefanos Vivas found 2,430 to be protectable and infringed. He said the court need not go beyond a set of headnotes, but also argued that headnotes are categorically copyrightable based on long-established case law.
Cendari then turned to fair use and framed the case as one of the alternatives. Citing Supreme Court authority, he said substitution is “copyright fraud” and that “this is a classic case of substitution” in that ROSS developed and sold a product to replace Westlaw.
She pointed to ROSS marketing materials that encourage users to “choose ROSS or Westlaw,” which she said is a classic example of a substitute product, according to Thomson Reuters.
Judge Bove asked Sendari to identify the relevant market with respect to the fourth element, noting that the parties appeared to be “at a distance” on this issue.
Cendali identified three forms of market damage. Damage to ROSS’s product competing as a replacement for Westlaw, damage to Thomson Reuters’ ability to use its proprietary Headnotes exclusively to train its own AI products, and damage to the potential market for other AI companies to license Headnotes for AI training.
Mr. Cendari disputed Mr. ROSS’s finding that Thomson Reuters did not use headnotes for AI. He said Thomson Reuters trained with its own headnotes years before ROSS was founded, and Westlaw’s use of AI and natural language search predates ROSS. He also pointed to a court hearing by Lexis as evidence that other legal research providers were developing AI products around the same time.
“So they try to portray themselves as great innovators, but in reality they were latecomers,” Sendari argued.
Judge Montgomery Reeves drew parallels with judges on the Third Circuit. video pipeline The Disney trailer case asks whether the lack of an independent market for the copied material trumps the market harm argument.
That was not the case, Sendari said. She claimed: video pipelinecopied content can undermine the value of the broader platform, even if it is not itself sold separately.
Regarding the first fair use element, Mr. Cendali argued that ROSS’s use is not transformative because it uses Westlaw headnotes to construct a competing legal research tool. Training is just a means to an end, she said, and under the Supreme Court’s ruling, warhol When making a decision, we will conduct an appropriate investigation focusing on the purpose of use. Here, both products served the same purpose of assisting users with their legal research.
Judge Bove asked what the record showed about the nature of ROSS’s training and skills. Cendali responded that the product would eventually return cases and estimates, similar to what Westlaw did. He distinguished this case from disputes involving generative AI systems that generate new documents, and said ROSS’s product is a legal research system for finding court opinions in response to legal topics.
Cendali also argued that ROSS took away the “heart” of Westlaw’s research, even though the number of headnotes copied was small compared to Westlaw’s entire database. She said Westlaw’s editorial enhancements, including headnotes and a key number system, distinguish it from just a casebook. Rather than creating her own questions from the underlying opinions, Ross took the question-and-answer pairs created through that editorial work, she said.
The Panel also investigated the relevance of cases involving intermediate copies such as: sega, sony and google vs oracle. Cendali argued that these incidents do not help ROSS because they involve copies of working computer code that are necessary to understand or achieve compatibility. Here, ROSS had access to public judicial opinions and didn’t have to copy Westlaw’s headnotes to build the system, she said.
Counterargument: “AI” is not just a label
In response, Davis returned to the technology and emphasized that ROSS is not just invoking “AI” as a label. He said the system uses deep neural network technology similar to systems later associated with ChatGPT and Anthropic, but acknowledged that ROSS’s product is not generative AI.
In response to Judge Montgomery Reeves, he agreed that ROSS did not prepare a brief or prepare a legal response, but instead returned quotes from the judicial opinion.
Mr. Davis also disputed Thomson Reuters’ market harm evidence, arguing that the record lacked concrete evidence that ROSS harmed Westlaw’s markets. Judge Bove reminded ROSS that, based on precedent, ROSS bears the burden of fair use, including Factor 4.
Davis acknowledged the burden, but argued that Thomson Reuters was better able to provide evidence such as subscriber numbers and customer data. He said the records include evidence that some users have left ROSS and returned to Westlaw.
The discussion ended with a debate over whether the intermediate copy case had any meaningful similarity. Judge Bove pointed out that: sega and sonyHumans cannot read object code, so a copy was required, but ROSS employees were able to read the judicial opinion and create their own training materials from it.
Davis responded that there is no strict necessity requirement for fair use, and that ROSS has a legitimate purpose as a startup to build new legal research technology under time pressure.
After arguments, Judge Bove looked at the issue more broadly, telling Davis, “I think this case is about balancing what we have and where are the lines that we shouldn’t cross.”
“As with all copyright litigation, it’s all about balancing,” Davis agreed. “Your Honor”
The court has not announced the timing of its ruling.
