Perhaps this is the way the case will end, not with a bang but a whimper.
With apologies to T. S. Eliot, that was my reaction when I read the 3rd U.S. Circuit Court of Appeals’ assessment of the 6.5-year battle between legal information giant Thomson Reuters and once-fledgling startup ROSS Intelligence as “no more than an ordinary copyright case.”
From the start, ROSS’s founders had portrayed this as a David-v-Goliath battle between a huge, monopolistic information company with unlimited resources aiming to shut down a competitor seeking to democratize the law through better legal technology.
ROSS also characterized the case as a referendum on the future of AI technology in law and beyond. But Judge Tamika R. Montgomery-Reeves, writing for the three-judge panel, rejected that framing.
“Under ROSS’s framing, this case appears to concern the future of AI legal technology,” she wrote. “But appearances can be deceiving. In truth, this is no more than an ordinary copyright case.”
The 3rd Circuit issued its opinion under seal on Sept. 29. Within a day, word emerged that the court had ruled for Thomson Reuters. The court released the full text of the opinion shortly after that.
The panel, which also included Judges L. Felipe Restrepo and Emil J. Bove, affirmed Judge Stephanos Bibas’s February 2025 grant of partial summary judgment to Thomson Reuters on 2,243 headnotes. The court held that Westlaw headnotes are copyrightable and that ROSS’s use of them to train its AI legal research tool was not fair use.
Although the decision is the first by a federal appeals court to address whether using copyrighted material to train an AI system is fair use, the court explicitly sought to limit its reach.
“ROSS’s AI was not a generative AI, meaning it would not create any new expression; it would only return text passages from preexisting judicial opinions,” the court said.
For that reason, the court distinguished cases that had found that training a large language model can be a transformative use under copyright law, including Bartz v. Anthropic PBC, 787 F. Supp. 3d 1007, 1014 (N.D. Cal. 2025), and In re OpenAI, Inc. Copyright Infringement Litig., No. 1:25-md-3143 (S.D.N.Y. Sept. 1, 2026).
“Unlike the AI models in Bartz and In re: OpenAI, ROSS’s AI platform cannot generate original expression, and the evidence here supports the opposite conclusion about transformativeness,” the court said.
What the court did decide involved three key conclusions that form the basis of the legal rulings:
- First, ROSS copied the headnotes to build a direct commercial substitute for Westlaw, which the court treated as decisive on the first and fourth factors for determining fair use.
- Second, ROSS did not need the headnotes, because it already had the underlying judicial opinions, and “ease is not a justification for copying.”
- Third, a market for licensing headnotes as AI training data is “rapidly developing,” and ROSS’s unlicensed copying usurped Thomson Reuters’s opportunity to enter it.
Background of the Case
Thomson Reuters sued ROSS in Delaware federal court in 2020, alleging copyright infringement and tortious interference with contract. ROSS, founded in 2014 by three University of Toronto computer science students after an IBM Watson competition, had built a search engine that answered plain-language legal questions by returning passages from judicial opinions. It drew on a bank of roughly 10 million opinions.
To train the system, ROSS hired LegalEase Solutions to write about 25,000 legal memos, each posing a question and offering four to six opinion passages ranked as great, good, topical or irrelevant. LegalEase and a subcontractor, Morae Global, used thousands of Westlaw headnotes to write the questions.
“ROSS had access to the underlying judicial opinions and could freely copy them to make the memos needed to train its AI,” the court wrote. “It chose not to do so because copying the headnotes offered an ‘easy’ way to create its training memos.”
ROSS never hid its goal, the court said. It ran ads comparing itself to Westlaw at prices “in line with” Westlaw’s, its executives hoped it would serve as a substitute, and some law firms switched.
Bibas, a 3rd Circuit judge sitting by designation as the trial judge, initially sent most of the case to a jury in 2023. In February 2025, he reversed course and granted partial summary judgment to Thomson Reuters on 2,243 headnotes, finding infringement and rejecting fair use.
He then certified two questions for interlocutory appeal: whether the headnotes and the Key Number System are original, and whether ROSS’s use was fair.
The 2,243 figure came from a batch of 2,830 headnotes that ROSS’s own expert said closely resembled the memo questions but differed significantly from the underlying opinions. Bibas compared each one and concluded that for 2,243 of them, no reasonable juror could find they were not copied.
ROSS shut down its platform in 2021, citing litigation costs.
Headnotes: A Low Bar for Creativity
In its Sept. 29 opinion, the court held that all 2,243 headnotes clear the “extremely low” originality bar set by Feist Publications v. Rural Telephone.
For each one, Montgomery-Reeves wrote, Westlaw’s editors decided which points of law were important enough to include and how to word them under Westlaw’s drafting rules. Each headnote had to stand alone without the opinion while accurately reflecting it. Those choices supply the “modicum of creativity” copyright requires.
The court found support in two Supreme Court decisions decided a century apart. In Callaghan v. Myers (1888), the Supreme Court held that a reporter could hold a copyright in headnotes even though the opinions themselves were not protected.
More recently, in Georgia v. Public.Resource.Org (2020), the court described Callaghan as upholding the reporter’s interest in headnotes and similar materials, so long as the author had no authority to speak with the force of law.
The panel rejected ROSS’s three counterarguments:
- A copyright in headnotes would give Thomson Reuters a monopoly over the law. “Not so,” the court wrote. “Headnotes are not law; judicial opinions are.” The opinions remain free to all.
- Headnotes are not copyrightable under the merger doctrine. The doctrine applies only in the “rare” case where an idea can be expressed in one or a few ways. The court cited its own 2019 decision in Silvertop Associates v. Kangaroo Manufacturing, which held that a banana costume could take many designs. “Just as a banana costume can take myriad designs, so too can headnotes have differing expressions,” the court said.
- Headnotes are insufficiently original for copyright protection. The court distinguished the 1998 2nd Circuit case of Matthew Bender v. West, which denied West a copyright in party names, parallel citations and similar information because industry conventions dictated those choices. Headnotes are “independently composed” and not bound by conventions.
In a footnote, the court declined to address Bibas’s suggestion that even headnotes quoting opinions verbatim could be original, because the 2,243 headnotes at issue do not copy opinion text word for word.
It also did not address the originality of the Key Number System because ROSS never briefed that issue, even though it was among the questions certified for appeal.
ROSS’s Use Was ‘Minimally Transformative’
In its analysis of fair use, the court relied heavily on the Supreme Court’s 2023 decision in Andy Warhol Foundation v. Goldsmith, which asks whether a secondary use has a purpose distinct from the original and weighs that against the use’s commercial nature. Where both works share the same purpose and the use is commercial, Warhol says the first factor cuts against fair use.
ROSS’s use was commercial, the court found, because it built a platform priced to compete with Westlaw for the same customers.
On transformativeness, the court acknowledged that ROSS “did not use the headnotes in precisely the same fashion as Thomson Reuters.” Training an AI program was an intermediate step that “arguably presents a slight degree of difference in use.”
But both companies used the headnotes to build legal research platforms that help users find responsive law. That shared “ultimate purpose” made ROSS’s use “minimally transformative, at best,” the court said.
The court rejected ROSS’s reliance on Authors Guild v. Google, the 2nd Circuit’s 2015 decision upholding Google Books. There, scanning books to create a search tool served a completely different function from reading them, and the tool may have driven users to buy the books.
ROSS’s tool, by contrast, did what Westlaw already does with its headnotes, and ROSS admitted it aimed to replace Westlaw, not send users to it.
The court also rejected ROSS’s argument from the “intermediate copying” cases: Google v. Oracle, Sega v. Accolade and Sony v. Connectix. In each, the court said, copying computer code was necessary to reach unprotected functional elements and make software interoperable.
ROSS faced no such necessity, the court said. It had the judicial opinions and could have used them to write its training memos. “Unlike necessity, ease is not a justification for copying,” Montgomery-Reeves wrote.
In a footnote, the court said that to the extent good faith still plays a role in the first factor, ROSS’s conduct weighed against it. The record showed ROSS tried to access Westlaw using a law-firm investor’s credentials despite notice that doing so violated Westlaw’s terms, that an employee posing as a sole practitioner inquired about an account, and that another used a student account while hiding his employment with a competitor.
Factors Two and Three: A Split
The second factor, the nature of the copyrighted work, was the only one to favor ROSS. The headnotes were published before ROSS used them, and they are “more factual than fictional” because they must accurately describe uncopyrightable opinions. But the court noted, quoting Authors Guild, that this factor “has rarely played a significant role” in fair use outcomes.
On the third factor, the amount and substantiality of the portion used, the court sided with Thomson Reuters. ROSS argued it took only 0.08 percent of Westlaw’s 28 million headnotes. The court answered that each headnote is a copyrightable work in its own right, so for each headnote taken, ROSS copied an entire work.
“Copying the headnotes was also not necessary to train ROSS’s AI, as the underlying judicial opinions were freely available,” the court said. “So, ROSS took more than necessary when it copied these headnotes without a justifying transformative purpose.”
A Market for Headnotes
As to the effect of the use on the market for the copyrighted work, the court found market harm in three ways.
Value of the headnotes. ROSS argued there is no standalone market for headnotes, so there could be no harm. The court turned to its own 2003 decision in Video Pipeline v. Buena Vista Home Entertainment, in which a website argued no one pays to watch movie trailers. The 3rd Circuit rejected that argument then because the statute looks to harm to the value of a work, not just its market.
“By copying the headnotes and using them to build its own competing legal-research platform, ROSS appropriated the headnotes’ value for finding and analyzing judicial opinions and diminished their value as a draw for users to Thomson Reuters’s legal-research platform,” the court said.
The legal research market. The court accepted Thomson Reuters’s position that the legal research platform market is also a relevant original market. Unrestricted copying of headnotes to build substitute research tools “would harm Westlaw’s position in that original market,” and ROSS offered no evidence to rebut that.
Licensing headnotes as AI training data. ROSS argued that there is no derivative market for headnotes. The court disagreed, finding that “the market for licensing headnotes as text to train AI is rapidly developing.” It pointed to evidence that Thomson Reuters already trains its own AI search products on headnotes. That Thomson Reuters had not licensed headnotes to others did not disprove a market exists, the court said. By copying without authorization, ROSS “usurped Thomson Reuters’s opportunity to enter that derivative market.”
The panel also rejected ROSS’s claim that its copying would benefit the public by increasing access to the law. The court noted that the underlying opinions were already free and that ROSS was charging for its service at prices comparable to Westlaw.
ROSS also argued that affirmance would halt AI development, but it presented “no evidence to support that broad claim,” the court said. ROSS also invoked national security, but offered no evidence connecting its legal research tool to national security.
What Happens Next
Despite my opening line, this case is not over.
Because it was an interlocutory appeal, the case will now return to the district court. Still pending there are Thomson Reuters’s tortious interference claim and any issues concerning headnotes beyond the 2,243 resolved on summary judgment.
However, before that happens, ROSS has indicated in a court filing that it will ask the full 3rd Circuit to reconsider the panel’s decision.
It has publicly said it will ultimately seek review by the Supreme Court.
So the case will carry on. And I will carry on in covering it.
(For full coverage and background on this case, see my archive of stories.)