The Third Circuit held on September 29, 2026 that 2,243 Westlaw headnotes are original enough for copyright protection and that ROSS Intelligence’s copying of them to train a legal-research AI was not fair use. The court affirmed partial summary judgment for Thomson Reuters in what appears to be the first federal appellate ruling on fair use in AI training.1
Background
Thomson Reuters and West Publishing sued ROSS in the District of Delaware in 2020, alleging copyright infringement and tortious interference with contract. ROSS had built a search engine that answered plain-language questions with passages drawn from roughly ten million uncopyrighted judicial opinions. To train it, ROSS engaged LegalEase Solutions, which produced about 25,000 memoranda pairing a legal question with four to six opinion passages graded from great to irrelevant. Drafters used thousands of Westlaw headnotes to frame the questions because the headnotes offered an easy way to do so.
Judge Stephanos Bibas, a Third Circuit judge sitting by designation, granted partial summary judgment on February 11, 2025, finding 2,243 headnotes copied and rejecting fair use; he assigned the first and fourth factors to Thomson Reuters and the second and third to ROSS.2 At the parties’ request, he certified two questions for interlocutory appeal under 28 U.S.C. § 1292(b): whether the headnotes and the Key Number System are original as a matter of law, and whether ROSS’s use of the headnotes was fair. Judges Restrepo, Montgomery-Reeves, and Bove heard argument on June 11, 2026. Judge Montgomery-Reeves wrote for the panel.
Analysis
Originality in editorial choice
The panel held that Thomson Reuters’ headnotes are original works entitled to copyright protection. Under Feist, a work needs only independent creation and a modicum of creativity, a bar the panel called extremely low. Westlaw editors decide which points of law matter and how to word them under criteria that include an 800-character guideline and a requirement that each headnote stand alone without the opinion. Those decisions supply the creative spark. The panel read Callaghan v. Myers and Georgia v. Public.Resource.Org as supporting copyright in reporter-prepared headnotes. It answered ROSS’s monopoly-over-the-law argument on the ground that headnotes are not law, declined to apply the merger doctrine because points of law can be expressed many ways, and distinguished Matthew Bender & Co. v. West Publishing Co., which denied protection to party names and parallel citations rather than to independently composed headnotes.
Commercial copying, not transformation
Copying an original work is not infringement if it qualifies as fair use—a statutory exception the panel weighed factor by factor and found largely unavailable to ROSS. On the first fair use factor (purpose and character of the use) the court found the use highly commercial and minimally transformative at best: both parties use headnotes to help users find responsive legal material, making the training an intermediate step toward the same ultimate purpose. The panel distinguished Authors Guild v. Google, where the search function served a different purpose and could drive book sales, and the intermediate-copying line of Google v. Oracle, Sega, and Connectix, where copying was necessary to reach unprotected functional code. ROSS had the opinions themselves and chose headnotes because they were easier. “Unlike necessity, ease is not a justification for copying,” the court wrote. A footnote here set aside the Justice Department’s statement of interest in the OpenAI multidistrict litigation, along with Bartz and Kadrey, because ROSS’s system cannot generate original expression. A second footnote added that ROSS’s attempts to reach Westlaw through investor and student credentials weighed against it, to the extent good faith still matters.
The second factor (nature of the copyrighted work) favored fair use slightly, because the published headnotes are more factual than creative. The third factor (amount and substantiality of the portion used), which the district court had given ROSS, came out the other way: with no transformative purpose and the opinions freely available, ROSS took more than it needed, and each headnote it took was an entire copyrighted work.
A derivative market for training data
The fourth factor (effect on the potential market) supplies a consequential holding. Its analysis turned on which markets the panel recognized. Taking the legal-research platform market as relevant, it found that widespread copying to build substitutes would harm Westlaw’s position. ROSS framed the market more narrowly, as one for headnotes sold as a standalone product, and argued that no such market exists. The panel assumed that framing and still found harm: under Video Pipeline, the law directs courts to the effect of the use on the value of the work, not only on the market however narrowly defined, and Thomson Reuters promotes its headnotes as a draw for Westlaw subscriptions.
The court then turned to derivative markets, which cover uses that a copyright owner would generally develop or license others to develop. It credited evidence that licensing headnotes as AI training text is rapidly developing and that Thomson Reuters already trains its own AI search products on them, concluding that ROSS usurped that opportunity. ROSS’s public-benefit and national-security arguments failed for want of evidence; the opinions themselves are free, and ROSS priced its platform in line with Westlaw.
Takeaways
- Factor four now carries circuit-level recognition that licensing text as AI training data is a cognizable derivative market, even where the owner has licensed to no one.
- The ruling’s reach into generative AI disputes is bounded by the panel’s own footnote, which distinguished systems that produce new expression from ROSS’s, which only returned existing opinion passages.
- The panel treated this as an ordinary copyright case and said incorporating AI does not by itself excuse infringement.
Endnotes
- Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc., No. 25-2153 (3d Cir. Sept. 29, 2026) (precedential opinion), available at https://www2.ca3.uscourts.gov/opinarch/252153p.pdf.
- Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc., No. 1:20-cv-613-SB (D. Del. Feb. 11, 2025) (revised summary judgment opinion), available at https://www.govinfo.gov/content/pkg/USCOURTS-ded-1_20-cv-00613/pdf/USCOURTS-ded-1_20-cv-00613-5.pdf.
