The court called it “an ordinary copyright case.”
For legal technology businesses, that description deserves attention. An AI product can introduce a new way to search without changing the legal questions surrounding the material used to build it.
In Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc., the U.S. Third Circuit affirmed partial summary judgment protecting Westlaw’s editorial headnotes and rejecting ROSS’s fair-use defence. The opinion was filed September 29, 2026, and made public September 30. Crucially, the dispute involved a non-generative legal research tool - not a general ruling on generative AI training.
What the Westlaw-ROSS ruling decided
The court distinguished judicial opinions from the original editorial work involved in producing headnotes. It found sufficient creativity in editors’ selection and presentation of legal points.
Competition also mattered: ROSS used the material to develop a commercial legal research platform serving a highly similar purpose to Westlaw. Its training process did not, by itself, make that use sufficiently transformative. The court also considered harm to Westlaw’s market and potential licensing opportunities.
That distinction makes the decision useful beyond its immediate parties. The difficult question for a developer is often not simply whether information concerns the law. It is which parts of a dataset represent someone else’s editorial contribution.
Why Canadian legal technology teams should pay attention
Canada has its own copyright framework. Section 29 of the Copyright Act permits fair dealing for research, private study, education, parody or satire. The Third Circuit’s application of U.S. fair use does not determine whether a particular use qualifies as fair dealing in Canada.
Analysis: For Canadian founders, law firms and investors, the practical value of the decision is as a prompt for better questions about data.
A product demonstration shows what a system can do. A data review asks how the company obtained the ability to do it. Both deserve attention before a firm builds a workflow around a supplier.
The following questions offer a useful starting point:
- What entered the system? Ask vendors to distinguish judgments, publisher summaries, annotations and other source material.
- How was it obtained? Request an explanation of licences, suppliers and collection methods.
- What uses were authorized? Access, storage, model training and redistribution should be addressed separately.
- What happens if rights are challenged? Examine the supplier’s contractual commitments and the firm’s operational alternatives.
These are suggested diligence questions, rather than requirements imposed on Canadian firms by this U.S. judgment.
A better way to discuss AI copyright risk
Analysis: “We use publicly available legal information” is an incomplete answer when a product combines several types of content. A buyer needs enough detail to understand which materials are public legal texts and which have been selected, rewritten or organized by another business.
Similarly, describing a product as innovative says little about its permissions. The commercial case for an AI tool and the copyright case for its inputs require separate explanations.
For lawyers advising technology companies, this creates an opportunity to bring legal review into product development earlier. Discussing data sources before a system becomes dependent on them gives a business more room to change suppliers, negotiate rights or redesign its approach.
What remains open
The Third Circuit expressly distinguished ROSS’s system from generative AI models capable of producing original expression. Its reasoning should therefore be read within the product, evidence and competitive relationship before it.
For legal AI buyers, the next useful question is concrete: Can the vendor explain its training material as clearly as it explains its features?
Primary source: Read the Third Circuit opinion, No. 25-2153
Related coverage: Apple Challenges US$5.7 Billion Patent Verdict Over iPhone Haptics.