The 3rd U.S. Circuit Court of Appeals ruled that Thomson Reuters did not face an unprecedented generative AI dispute, categorizing a six-and-a-half-year copyright battle over Westlaw headnotes as an ordinary infringement case. The court affirmed a partial summary judgment against defunct startup ROSS Intelligence.
What began as a David-versus-Goliath startup narrative ended in a federal appeals court with a starkly traditional reading of copyright law. The 3rd U.S. Circuit Court of Appeals unsealed its ruling on September 29, 2026, affirming a February 2025 decision by Judge Stephanos Bibas that granted partial summary judgment to legal information giant Thomson Reuters. Writing for the three-judge panel alongside Judges L. Felipe Restrepo and Emil J. Bove, Judge Tamika R. Montgomery-Reeves dismantled the framing pushed by the defense.
Judge Tamika R. Montgomery-Reeves wrote that under ROSS’s framing, this case appears to concern the future of AI legal technology, but appearances can be deceiving, as in truth, this is no more than an ordinary copyright case.
The litigation originated in Delaware federal court in 2020 when Thomson Reuters sued ROSS for copyright infringement and tortious interference with contract. Founded in 2014 by three University of Toronto computer science students following an IBM Watson competition, ROSS built a search engine designed to answer plain-language legal questions using passages drawn from roughly 10 million judicial opinions. To train its system, ROSS hired LegalEase Solutions to draft approximately 25,000 legal memos. LegalEase and its subcontractor, Morae Global, used thousands of Westlaw headnotes to formulate the questions for those training memos.
Originality of Westlaw Headnotes and the Feist Standard
At the center of the dispute stood 2,830 headnotes identified by ROSS’s own expert as closely resembling the memo questions used in training. Trial judge Bibas compared each item and concluded that for 2,243 of them, no reasonable juror could find they were not copied. On appeal, the 3rd Circuit evaluated whether those 2,243 headnotes met the legal threshold for copyright protection.
An AI rival to Westlaw lost its fair-use appeal. The ruling does not settle every AI-training case.The Third Circuit said ROSS used Westlaw summaries to build a substitute research tool and expressly distinguished generative AI.https://t.co/ydVcGQeyci
The court held that the headnotes easily cleared the extremely low originality bar set by Feist Publications v. Rural Telephone. Each headnote reflects a specific point of law extracted from a judicial opinion by editors following editorial guidelines, embodying enough creative spark in both the selection of legal principles and the specific wording used to convey them. The appellate panel rejected three primary counterarguments raised by ROSS:.
- Monopoly on the law: The court emphasized that headnotes are created by editors, whereas judicial opinions remain free for anyone to publish.
- The merger doctrine: Rejecting the claim that headnotes are not copyrightable under the doctrine, the panel rejected ROSS’s counterargument.
- Insufficient creativity: The court distinguished earlier precedents such as Matthew Bender, concluding that the Westlaw headnotes carry distinct creative authorship.
Fair Use Analysis and the Rejection of the Transformative Defense
The legal battle hinged heavily on the four statutory fair use factors under U.S. copyright law. While the second factor—the nature of the copyrighted work—tipped in favor of ROSS because headnotes lean toward factual reporting rather than pure fiction, the remaining factors decisively favored Thomson Reuters Enterprise Centre GmbH.
On the first factor, examining the purpose and character of the use, the court determined that ROSS operated a highly commercial enterprise intended to compete directly with Westlaw. The panel distinguished the startup’s system from generative artificial intelligence, noting that Ross’s AI is not generative AI. Instead of writing new content, the platform retrieved pre-existing judicial opinions in response to user queries. Because ROSS used the headnotes to build a direct commercial substitute for Westlaw, the court found the usage minimally transformative at best.
The 3rd Circuit also dismissed comparisons to transformative search tools like Google Books, noting that while scanning books to index them served a distinct function, ROSS sought to provide the exact service Westlaw already delivered. Oracle, Sega v. Accolade, and Sony v.
ROSS had access to the underlying judicial opinions and could freely copy them to make the memos needed to train its AI. It chose not to do so because copying the headnotes offered an ‘easy’ way to create its training memos.
The 3rd U.S. Circuit Court of Appeals, 3rd U.S. Circuit Court of Appeals
On the third factor, addressing the amount and substantiality of the material used, the court rejected the defense argument that ROSS took only 0.08 percent of Westlaw’s 28 million headnotes. Because each headnote constitutes a distinct copyrighted work, taking an individual headnote meant copying an entire work without justification. For the fourth factor, the court identified multiple forms of market harm, noting that ROSS not only competed directly with Westlaw’s core platform but also usurped potential derivative licensing markets where Thomson Reuters could license its headnotes as AI training datasets.
Litigation Costs and the Shutdown of ROSS Intelligence
The operational reality of the six-and-a-half-year legal fight caught up with the startup long before the appellate ruling. ROSS shut down its platform in 2021, explicitly citing mounting litigation expenses. Throughout the proceedings, the company marketed its services at prices comparable to Westlaw while running advertisements positioning itself as an alternative to the incumbent provider.
While industry observers have scrutinized the decision for its potential influence on broader machine learning disputes, the appellate court explicitly walled off its reasoning from generative AI policy debates unfolding elsewhere. The panel noted in a footnote that its analysis did not address separate departmental filings concerning large language models, such as a Statement of Interest filed by the United States Department of Justice in the Southern District of New York. Legal representatives for ROSS have indicated that the company plans to seek review by the Supreme Court of the United States.