Quick Answer
The Third Circuit held that ROSS Intelligence’s copying of Westlaw headnotes was not fair use, and it affirmed partial summary judgment for Thomson Reuters. The September 29, 2026, ruling turns on the headnotes’ original editorial choices and ROSS’s commercial competing purpose. Legal researchers and AI developers should distinguish protected editorial summaries from freely available court opinions before building training datasets.
Key Takeaways
- The Third Circuit affirmed partial summary judgment for Thomson Reuters in its case against ROSS Intelligence.
- The appeal involved 2,243 Westlaw headnotes that the court found sufficiently original for copyright protection.
- ROSS copied the full text of 25,000 Westlaw-written headnotes, according to the appellate opinion.
- The court found ROSS’s use highly commercial and minimally transformative because it sought to compete with Westlaw.
- The ruling concerns a legal research platform, not a generative AI chatbot, but it provides a significant appellate analysis of training-related copying.
What did the AI training copyright ruling decide?
The Third Circuit ruled that ROSS Intelligence’s use of Westlaw headnotes was not fair use and affirmed partial summary judgment for Thomson Reuters. The September 29, 2026, decision in Thomson Reuters Enterprise Center GmbH v. ROSS Intelligence Inc., No. 25-2153, concluded that the Westlaw headnotes at issue had enough original expression to receive copyright protection.
The appellate court’s conclusion matters because ROSS did not simply use the underlying judicial opinions. The court explained that judicial opinions were freely available, while the disputed material consisted of editorial headnotes written by Westlaw. Readers evaluating an AI training copyright ruling should separate public source material from an editor’s protected selection and phrasing.
The court’s decision does not establish that every use of material in an AI dataset is infringing. Fair use depends on the specific work, the purpose of the use, and the other statutory factors. The practical lesson from this case is narrower but important: copying proprietary editorial content to create a competing research product faces a difficult fair use argument.
Why did the court say Westlaw headnotes qualify for copyright protection?
The Third Circuit held that the 2,243 Westlaw headnotes in the appeal were sufficiently original because each reflected editorial choices. The court said Westlaw editors made decisions about which legal points to include and how to express those points, giving each headnote the required “creative spark.”
Westlaw headnotes are short summaries that identify legal issues and organize points from court opinions. The underlying opinions remain available to the public, but the court treated the headnotes as separate editorial expression rather than a purely mechanical reproduction of judicial text. The full appellate reasoning appears in the Third Circuit’s published opinion.
The distinction is central for developers and organizations that collect legal materials. A database can contain both public domain source documents and proprietary editorial layers, such as summaries, classifications, annotations, and editorial taxonomies. The most sensible approach is to identify those layers before data is copied, licensed, or used for model development.
| Material discussed in the case | How the Third Circuit described it | Why the distinction mattered |
|---|---|---|
| Judicial opinions | Freely available source material | The court did not treat public court opinions as the protected content at issue. |
| Westlaw headnotes | Editorial writing with selection and phrasing choices | The court found the headnotes sufficiently original for copyright protection. |
| ROSS’s copied material | The full text of 25,000 Westlaw-written headnotes | The volume and purpose of the copying informed the fair use analysis. |
Why did the court reject ROSS Intelligence’s fair use defense?
The Third Circuit rejected ROSS Intelligence’s fair use defense because the court found the use highly commercial and minimally transformative. ROSS intended to build a commercial legal research platform that would compete with Westlaw, according to the appellate decision.
Fair use analysis does not turn on a single factor, and commercial activity alone does not end the inquiry. In this dispute, however, the court viewed the relationship between the copied headnotes and ROSS’s planned product as especially important. ROSS used Westlaw’s editorial work to help create another legal research service rather than using the headnotes for criticism, commentary, scholarship, or a materially different expressive purpose.
The court also considered that ROSS copied entire headnotes rather than limited excerpts. That finding matters because the case involved the complete text of 25,000 Westlaw-written headnotes. Organizations considering training data should document why material is used, what portion is collected, and whether a product could substitute for or compete with the original service.
What material did ROSS Intelligence copy for its legal research platform?
ROSS Intelligence copied the complete text of 25,000 Westlaw-written headnotes, while the underlying court opinions were freely available. The Third Circuit treated that difference as a key part of the dispute because ROSS could access judicial opinions without copying Westlaw’s editorial summaries.
The case did not involve a consumer chatbot producing answers from a broad web corpus. ROSS was developing a legal research platform, and the court focused on the relationship between the copied headnotes and the planned competing service. Reuters described the decision as the first U.S. appeals court ruling in the current wave of AI training copyright disputes, while also noting that ROSS’s product was a legal search system rather than generative AI. Reuters’ report on the ruling provides that broader context.
The practical boundary is not simply whether software uses machine learning or automation. The more relevant questions include what was copied, whether the content contains protected expression, and whether the resulting product uses that expression to compete with the source. Publishers seeking technical options to state their preferences can also examine AI training controls for websites, although technical controls do not replace a legal analysis of a particular dataset.
Does the Westlaw ruling apply directly to generative AI models?
The Westlaw ruling does not directly decide whether training a generative AI model on books, news articles, images, code, or public web pages is fair use. The Third Circuit decided a dispute about Westlaw headnotes and ROSS’s legal research platform, so the court’s analysis is tied to those facts.
The decision still matters to generative AI disputes because it addresses copyrightability, copying, commercial purpose, transformation, and market competition. Those are recurring issues in litigation over AI training. A future court considering a different model, a different dataset, or a different output system could weigh those facts differently.
For most readers, the careful interpretation is that the ruling supplies a precedent, not a universal answer. The decision supports the view that proprietary editorial content receives meaningful protection when copied to build a competing product. It does not determine the outcome of every lawsuit involving machine learning, retrieval systems, or generative AI.
Questions about AI accountability also extend beyond copyright. The separate issues in an AI agent consumer-risk investigation concern potential consumer harm rather than ownership of training data, which is why the legal questions should not be treated as interchangeable.
What does the ruling mean for AI training datasets?
The Third Circuit ruling means AI training datasets require closer review when they include proprietary summaries, annotations, classifications, or other editorial material. A dataset may include public facts or public documents while also including protected expression added by a publisher, researcher, or database operator.
Data provenance is important because a developer needs to know where a dataset came from, what rights attach to it, and whether it contains material beyond the public source record. In the Westlaw dispute, the difference between the freely available judicial opinions and Westlaw’s authored headnotes shaped the court’s analysis. The practical response is to maintain records of source material, licenses, transformations, and dataset access restrictions before a model is trained.
Organizations should also avoid treating an internal label such as “publicly available” as proof that reuse is permitted. Public access and copyright status are different questions. A work can be easy to obtain online while retaining copyright protection in its original editorial expression.
Individual users do not need to resolve these legal questions when using ordinary AI tools, but they should remain cautious with sensitive material. People who use consumer chatbots can review chatbot privacy and data retention practices before submitting personal, financial, or confidential information.
What happened to ROSS Intelligence after the Thomson Reuters lawsuit?
ROSS Intelligence shut down its platform in 2021, citing the cost of the litigation after Thomson Reuters sued the company in 2020. The Third Circuit’s September 2026 ruling affirmed partial summary judgment for Thomson Reuters and upheld the conclusion that ROSS could not use Thomson Reuters content to build a competing platform.
Judge Stephanos Bibas wrote the appellate decision. The Associated Press reported that the ruling affirmed the finding that ROSS could not use Thomson Reuters content to create a competing service, a conclusion that places the dispute’s commercial purpose at the center of the decision. The Associated Press account of the case identifies Bibas as the author of the appellate opinion.
The immediate practical result is clear for the parties: Thomson Reuters retained its partial summary judgment victory, and ROSS is no longer operating its legal research platform. For the broader AI industry, the decision is a carefully fact-specific appellate ruling that will likely be studied in later disputes involving proprietary data and competing AI products.
How should developers and publishers respond to the AI training copyright ruling?
Developers should review training and retrieval datasets for proprietary editorial content before using them in commercial products. The Westlaw decision shows why a review should distinguish original documents from value-added content such as headnotes, summaries, metadata, annotations, and curated legal classifications.
- Identify the original source documents and any editorial material added by a database, publisher, or service.
- Review licenses, contracts, access terms, and internal records showing how the material was obtained.
- Document the proposed product purpose, especially if the product could compete with the source service.
- Limit use or seek permission when a dataset contains proprietary expression that is not necessary for the intended function.
- Consult qualified copyright counsel before launching a product based on disputed or licensed data.
Publishers should maintain clear records showing how their editorial products differ from underlying public materials. The court’s analysis gave weight to Westlaw editors’ choices about which legal points to include and how to phrase them. Clear documentation can help explain where protected editorial work begins.
Developers should stop and seek legal counsel before relying on fair use to justify copying a competitor’s proprietary database or editorial content. Copyright analysis is fact-specific, and the cost of resolving a dispute can remain substantial even when a company believes its use is defensible.
FAQ
What did the Third Circuit decide in the Thomson Reuters case?
The Third Circuit held that ROSS Intelligence’s use of Westlaw headnotes was not fair use and affirmed partial summary judgment for Thomson Reuters. The court found the headnotes sufficiently original for copyright protection.
Were the underlying court opinions copyrighted in this case?
The underlying judicial opinions were freely available, according to the Third Circuit opinion. The dispute focused on Westlaw-written headnotes, which the court treated as separate editorial expression.
How many Westlaw headnotes did ROSS copy?
ROSS copied the complete text of 25,000 Westlaw-written headnotes, according to the appellate opinion. The appeal itself concerned 2,243 headnotes that the court evaluated for copyright protection.
Does the ruling decide all AI training copyright lawsuits?
No, the Westlaw ruling does not decide all AI training copyright lawsuits because it concerns a legal research platform and specific Westlaw editorial content. Other cases can involve different datasets, products, uses, and fair use arguments.
What should AI developers do after the Westlaw headnotes decision?
AI developers should review datasets for proprietary editorial material and preserve records showing the source and rights status of training data. Developers should seek qualified legal advice before using material from a competitor’s paid database or other disputed source.