Third Circuit Rejects Fair Use Defense in Landmark AI Legal Research Copyright Case - FindLaw

The Third Circuit upheld Thomson Reuters’ copyright win, finding that Ross Intelligence’s use of Westlaw headnotes to train a competing AI legal research tool was not fair use. The ruling is a major early appellate decision on AI training, though the holding specifically distinguishes this case from generative AI training.

Judge Stephanos Bibas admitted he’d made a mistake when considering a motion for summary judgment in a copyright infringement case. The US Court of Appeals for the Third Circuit agreed that he got the ruling correct the second time.

On September 29, 2026, a three-judge Third Circuit panel affirmed a 2025 lower court decision in Thomson Reuters Enterprise Centre GmbH (TR) and West Publishing Corp v. Ross Intelligence Inc., a landmark copyright dispute ruling concerning artificial intelligence (AI), the fair use defense, and legal search tools. Judge Bibas ultimately ruled that Ross’s use of Westlaw editorial material to train its AI legal-research system was not transformative enough to qualify as fair use. Ross later pursued an interlocutory appeal, even though it had shut down in 2021 from litigation costs.

The Third Circuit rejected Ross’s arguments that Westlaw’s headnotes lacked sufficient originality for copyright protection and that Ross’s copying was fair use. Instead, the court held that the headnotes were copyrightable and that Ross used them commercially to help build a competing legal-research product.

How Ross Trained Its AI Legal-Research Tool

TR’s Westlaw subscription-based computer-assisted legal research service and database has long been a leader in the legal realm. In addition to its West Key Number System and a virtual library of court decisions, statutes, administrative regulations, and secondary legal sources, it boasts over 28 million headnotes. Although they can’t be cited as binding legal precedent in court, the headnotes summarize the most important legal points and holdings for each ruling, law, and case. Westlaw legal editors who create the headnotes follow strict guidelines, such as staying under 800 characters and providing enough factual background and reasoning while standing independent of other headnotes or the opinion itself.

Ross Intelligence was a legal-tech startup building an AI-powered legal-research platform. Unlike a generative-AI chatbot, Ross’s system responded to plain-language legal questions by returning relevant passages from existing judicial opinions. Ross Intelligence attempted to purchase access to the online data of TR's subscription-based Westlaw legal research platform. They intended to use Westlaw to train their own AI legal research system. TR threw a wrench into those plans by refusing to allow a potential future competitor to use Westlaw against them.

Ross hired LegalEase Solutions to create that training data. LegalEase and a subcontractor prepared about 25,000 “Bulk Memos,” which included Westlaw's headnotes. Ross converted the resulting memos into machine-readable data to train its AI system

Judge Bibas initially declined to grant either side summary judgment in 2023. But in February 2025, he granted Thomson Reuters partial summary judgment, holding that Ross had infringed copyrights in 2,243 Westlaw headnotes and that Ross’s use was not fair use. At the parties’ request, the court certified the key originality and fair-use questions for interlocutory review, leading to Ross’s appeal.

Out of Business, Out of Luck

In upholding the district court’s ruling, the Third Circuit addressed each facet of Ross Intelligence’s claim that it was engaging in fair use. For the originality argument, it agreed with Judge Bibas that in “2,243 headnotes, the memo questions were so similar to the headnote text but so dissimilar to the opinion text that no reasonable juror could conclude these headnotes were not copied.” It held that the headnotes Westlaw’s editors created were sufficiently original to qualify for copyright protection and, because they weren’t judicial opinions, wouldn’t substantiate Ross Intelligence’s assertion that they gave TR and Westlaw “a monopoly over the law.”

The panel also determined that, under 17 U.S.C. § 107, three of the four deciding factors indicated that Ross Intelligence’s use of the bulk memos was not fair use of Westlaw’s data. Although Ross Intelligence argued it used only about 0.08% of Westlaw’s headnotes for training, the court rejected that defense because Ross copied entire headnotes and took more than was necessary for a “minimally transformative” purpose.

As is often the case with changing technology, the judiciary has been cautious and meticulous in adapting to AI's ever-growing effects. The appellate court's confirmation of the district court’s ruling may provide future cases a precedential framework.