June 11, 2026, 21:23 GMT | Comment The question of whether Westlaw headnotes authored by Thomson Reuters attorney-editors are sufficiently creative to warrant copyright protection and — if so — whether their use by Ross Intelligence to develop an artificial intelligence-powered legal research tool qualifies as fair is now before the US Court of Appeals for the Third Circuit.

In Philadelphia on Thursday, Judges L. Felipe Restrepo, Tamika R. Montgomery-Reeves and Emil J. Bove seemed prepared to directly tackle the copyright implications of AI training — largely steering clear of efforts by Ross to redirect their attention to the threshold question of originality.

That the clash relates to nongenerative AI has done little to dampen interest from both the tech and creative sectors: more than 20 amicus briefs are on file in Ross’s effort to undo February 2025 findings (see here) that it's an infringer that can't evade liability under Section 107 of the Copyright Act.

The appeal arrives amid uncertainty over where copyright law draws the line between protected editorial expression and uncopyrightable legal or factual material — a question that remains unsettled even after Supreme Court intervention. As recently as April, the Third Circuit, in a precedential opinion authored by Restrepo, grappled with related questions involving copyright protections for standards incorporated into law in ASTM v. UpCodes (see here).

Still, the packed Albert Branson Maris Courtroom — where just a handful of seats went unfilled Thursday — was almost certainly attributable to the dispute’s sizable AI component, and the large-looming question of whether unauthorized use of copyrighted works in training is fair.

— ‘They’re the same’ —

At fair-use factor one — the purpose and character of the use — the panel pressed White & Case’s Mark Davies to explain how Ross’s reliance on Westlaw headnotes could be considered transformative.

In response, he said the underlying materials were converted into training data used to teach a machine-learning system “how to think like a lawyer,” stressing that users never saw the copied headnotes.

Bove then questioned whether Ross’s platform was readily distinguishable from existing legal research tools. “I'm sincerely trying to understand,” the judge said, later asking Davies to explain “what was so different or transformative or earth-shattering about Ross” compared to Westlaw.

After several exchanges, Davies acknowledged that, “from a user perspective, if you're asking to find a case . . . at some level, they're the same.”

He nevertheless maintained that the inner workings of Ross’s platform differed fundamentally from Westlaw in that Ross used deep-learning techniques and neural networking to connect legal questions with relevant judicial opinions, presenting the tool as emblematic of “the fundamental AI that is changing the world today.”

The panel appeared skeptical of his reliance on Sega v. Accolade and Sony v. Connectix Corp. in which the Ninth Circuit in 1992 and 2000, respectively, found certain forms of intermediate copying during software development permissible. Bove noted that the cases involved reverse engineering of computer code — whereas Ross already had access to the judicial opinions underlying Westlaw headnotes, but chose not to use them.

Thomson Reuters counsel Dale Cendali of Kirkland & Ellis urged the court to focus on purpose rather than technological implementation.

Ross’s stipulation that its platform was commercial and that it competed directly with Westlaw fits comfortably within the “same or highly similar purpose” framework for weighing factor one set by the Supreme Court in 2023’s Andy Warhol Foundation v. Goldsmith, she said.

— Commercial substitute —

Also garnering significant interest Thursday was fair use factor four, impact on the market of a work, which weighs considerations like a copyright owners’ lost sales and market dilution against the public benefit of the allegedly infringing product.

“It seems like there’s a distinct legal question about what market factor four asks us to think about,” said Bove.
Davies argued there is no market for headnotes for AI training, making the factor irrelevant, but Cendali pointed to Westlaw’s own forays into the AI space as evidence to the contrary.

Davies also zoomed out to argue that copyright “is not about protecting markets,” adding, “we want competition in this country” and “the Copyright Act is not an anti-competitive act.”

But Bove nevertheless pressed him to compare Westlaw to Ross’s “ultimate product,” suggesting an interest in market dilution and substitution shared by the entire panel.

“We would be competing with them,” Davies said. “I think that’s what you’re getting at.”

Montgomery-Reeves was more explicit.

Ross “aimed to serve as a commercial substitute for Westlaw,” she said, pointing to the appellant’s own marketing materials in support.

Cendali also referenced Ross’s advertising strategy in her arguments, presenting it as a “quintessential” example of marketing a substitute product. What’s more, she continued, those actions came at a steep cost for Thomson Reuters — depriving the publisher of not only subscribers but also potential licensing revenue for AI training data.

The second and third fair-use factors — the nature of the copyrighted work and the portion used — garnered less attention at Thursday’s hearing.

The Westlaw headnotes are “not like poetry,” Davies said, arguing that his client should win on the second factor. But Montgomery-Reeves countered that the bar for originality “doesn’t seem very high.”

“I think everybody agrees this is not some remarkable level of creativity,” Bove added. “But there are things going on here that go beyond copying.”

More importantly, the judges said factors two and three carry less weight in the fair use analysis than factors one and four.

“If you take two and three and they take one and four, you win?” Restrepo asked incredulously.

“It’s a multi-factor test,” Davies replied.

— Observable discretion —

Although it did not drive Thursday’s hearing, the Third Circuit touched on Ross’s position that Westlaw headnotes aren't entitled to protection because they merely restate judicial holdings — works which, by law, are uncopyrightable under the government edicts doctrine.

Davies stressed the lack of creative expression, noting that Westlaw attorney-editors are instructed to track judicial language as closely as possible. “Trivial” deviations are too slight to satisfy copyright law's originality requirement, he argued.

But the panel was unconvinced that the appellee’s editorial process is entirely mechanical.

Montgomery-Reeves pointed to examples in which attorney-editors condensed lengthy passages from judicial opinions into shorter statements of legal principles and questioned why such work is not “sufficiently original.” Bove, likewise, was skeptical of Ross’s position.

Referring to a headnote based on the Virginia Supreme Court's 1953 decision in Seymour & Burford Buick Corp. v. Richardson, he observed that the attorney-editor had omitted citations and other portions of the opinion and made independent judgments about what information to highlight in the corresponding headnote.

Even in an example Ross described as a “straight copy,” Bove said, “there is discretion.”

Davies countered that copyright can't arise merely because an editor identifies or restates legal propositions found in a judicial opinion. To hold otherwise, he warned, would wrongly remove portions of the law from the public domain.

According to Cendali, the issue before the court is narrower than Davies suggests. The summary judgment on appeal, she noted, included an assessment of thousands of Westlaw headnotes that not only differed from the underlying opinions but also were admitted, by Ross's own expert, to be non-verbatim copies.

The Supreme Court has long drawn a distinction “between the work of people writing about law, and the law itself,” she continued, and Thomson Reuters’ headnotes are no exception.

"The Constitution is designed to give people copyright protection so that they can innovate . . . and that's what my client has done for over 100 years," Cendali said.

Whether the panel ultimately affirms or revives Ross's defenses, Thursday's argument suggested the court is unlikely to avoid the fair use issues that have hovered over AI-training litigation for years. Any opinion could become the first appellate roadmap for district courts now confronting similar questions in more than 100 copyright cases against AI developers.

Please email editors@mlex.com to contact the editorial staff regarding this story, or to submit the names of lawyers and advisers.