US Appeals Court Upholds Thomson Reuters’ Landmark Win in AI Training Lawsuit

A U.S. appeals court has upheld a landmark copyright ruling in favor of Thomson Reuters, rejecting Ross Intelligence’s argument that it could use copyrighted Westlaw material to train an AI-powered legal research tool under the fair-use doctrine.

The Third U.S. Circuit Court of Appeals ruled on September 29, 2026, affirming a lower-court decision that Ross's use of Thomson Reuters' copyrighted Westlaw headnotes was not fair use. It is the first ruling from a U.S. appeals court addressing whether the use of copyrighted material to train an AI system can qualify as fair use.

The decision could become an important reference point as publishers, technology companies, authors and other copyright owners continue fighting over how protected content can be used to develop artificial intelligence.

What Happened in the Thomson Reuters AI Lawsuit?

The dispute dates back to 2020, when Thomson Reuters sued Ross Intelligence, a legal-research startup that was developing an AI-powered search product designed to compete with Westlaw.

Thomson Reuters alleged that Ross had copied thousands of Westlaw headnotes to help train its system. Headnotes are editorial summaries of legal points contained in court opinions. While the underlying judicial opinions are generally public, Thomson Reuters' editorial work in creating Westlaw's headnotes and its organizational system can contain copyrightable expression.

Ross had initially sought a license to Thomson Reuters' content, but Thomson Reuters declined. Ross subsequently used legal-service provider LegalEase to create "Bulk Memos" containing legal questions and answers derived from Westlaw headnotes, according to the lower court record.

The resulting dispute became an early test of whether intermediate copying of copyrighted material for AI training could be protected by fair use.

Appeals Court Rejects Ross' Fair-Use Argument

The Third Circuit's decision leaves intact the lower court's February 2025 ruling.

Judge Stephanos Bibas, sitting by designation in the Delaware federal court, had ruled that Ross infringed Thomson Reuters' copyrights and rejected Ross' fair-use defense. The court concluded that the use was commercial and non-transformative and that the competing legal-search purpose affected the potential market for Thomson Reuters' content.

The appeals court affirmed that result.

However, there is an important limitation: the Third Circuit's detailed reasoning is currently sealed. That means the public does not yet have the full appellate explanation of how the judges reached their conclusion.

That makes the ruling significant, but it is too early to treat it as a comprehensive legal rule covering every form of AI training.

Why the Case Matters for AI Companies

The central issue is not simply whether AI companies can train models on copyrighted works.

The more important question is under what circumstances they can do so without permission.

AI systems are trained using large quantities of information. Depending on the system, that information can include books, articles, images, code, legal material and other copyrighted works.

Technology companies have argued in various cases that copying material during the training process can be legally permissible because the material is used to develop a new technological system rather than being reproduced directly for users.

Copyright owners, meanwhile, have argued that AI companies should not be able to commercially exploit protected content without authorization or compensation.

The Thomson Reuters case adds weight to the argument that commercial AI development does not automatically make intermediate copying transformative or fair.

But the facts of this case are unusually specific.

This Was Not a Generative AI Case

One of the most important details for investors and technology companies is that the Ross system was not a generative AI model.

Ross was developing an AI-based legal search engine that returned judicial opinions in response to legal queries. It did not generate answers in the same way that modern large language models produce text.

That distinction limits how broadly the ruling can be interpreted.

Generative-AI lawsuits involving books, news articles, images, music and other content often involve different technologies, datasets, licensing arrangements and outputs.

The Third Circuit's decision therefore does not establish that all AI training using copyrighted material is unlawful.

Instead, it provides an important appellate-level data point in a rapidly developing area of copyright law.

The Market for AI Training Data Is Becoming More Important

The ruling comes at a strategically important moment for Thomson Reuters itself.

The company is increasingly positioning proprietary professional information as an asset for AI products. In August 2026, Thomson Reuters launched Thomson, its own large language model trained on Westlaw, Practical Law, Checkpoint and Reuters content. The model is being used within CoCounsel Legal.

That development makes the economics of proprietary content particularly relevant.

High-quality legal, tax and compliance information is difficult to reproduce because its value is not simply the raw information. Editorial organization, expert validation, structured databases and trusted workflows can all contribute to the commercial value of the content.

If courts strengthen the ability of copyright owners to control commercial use of that material for AI development, proprietary datasets could become strategically more valuable.

What It Could Mean for Thomson Reuters

For Thomson Reuters, the ruling potentially reinforces the importance of its intellectual-property portfolio and proprietary professional content.

The company has been investing heavily in AI products designed around its legal, tax, accounting and compliance databases. Its strategy increasingly combines proprietary information with AI rather than competing solely as a traditional information publisher.

That creates two potentially important effects.

First, copyright protection can strengthen the commercial value of licensed and proprietary datasets.

Second, the legal precedent could make it more difficult for competitors to replicate specialized AI products by using protected third-party content without permission.

That does not eliminate competition. AI developers can still build systems using licensed data, public-domain information, independently created datasets and other legally permissible sources.

Potential Impact on AI Developers

For AI companies, the ruling could increase the importance of data provenance and licensing.

Data provenance refers to understanding where training data came from and what rights exist to use it.

A company that can demonstrate that its training material was properly licensed or otherwise lawfully obtained may have a different legal position from a competitor accused of copying protected material without authorization.

That could encourage more licensing agreements between AI companies and publishers.

It may also increase the cost of developing specialized AI models where high-quality proprietary data is essential.

For startups, those costs could be particularly significant because licensing premium datasets can require substantial capital.

What Investors Should Watch Next

The most important issue now is how courts apply copyright law in other AI-training cases.

Investors should watch several areas:

1. Other appellate rulings: The Thomson Reuters case is the first U.S. appellate ruling on AI-training fair use, but other courts are considering different factual scenarios.

2. Generative AI cases: Future decisions involving large language models and creative content could provide more direct guidance for companies such as major AI developers.

3. Licensing deals: If legal uncertainty encourages AI companies to license books, news, legal databases and other content, publishers with valuable proprietary datasets could gain a new revenue opportunity.

4. AI costs: More licensing requirements could raise the cost of training specialized models and change the economics of AI development.

5. Thomson Reuters' AI growth: Investors will want to see whether proprietary content translates into higher adoption and recurring revenue from products such as CoCounsel Legal.

A Landmark Decision, but Not a Final Answer

The Third Circuit's ruling is important because it moves the AI-copyright debate from largely unsettled trial-court territory into the appellate courts.

But the decision should not be interpreted as a blanket prohibition on training AI with copyrighted material.

The case involved specific copyrighted Westlaw headnotes, a competing legal-search product, a particular method of obtaining training material and a fair-use defense. The appellate court's full reasoning remains sealed.

Those limitations will matter as courts consider very different AI-training disputes.

Conclusion

The U.S. Third Circuit's decision to uphold Thomson Reuters' victory over Ross Intelligence marks an important development in the legal battle over AI training and copyright.

The immediate result strengthens Thomson Reuters' position over its Westlaw content and provides the first U.S. appellate ruling rejecting a fair-use defense involving AI training. At the same time, because the case involved a non-generative legal-search system and the appellate reasoning remains sealed, its broader impact will depend on how other courts apply copyright law to different AI technologies and datasets.

For investors, the bigger theme is the growing economic importance of proprietary, legally controlled data as AI becomes embedded in professional software.

Follow the blog for more AI, technology, business and market developments.

This article is for informational and educational purposes only and should not be considered investment advice.