A defunct artificial intelligence company argued before a three-judge Third Circuit panel that its unauthorized use of Westlaw data to train a competing legal research tool qualifies as fair use. The outcome will determine how strictly courts apply copyright law to AI training data in the legal sector, potentially setting a precedent for how proprietary legal databases protect their editorial content.
The fair use dispute
Thomson Reuters sued ROSS Intelligence in 2020 after the startup hired a third party, LegalEase Solutions, to scrape Westlaw headnotes. Reuters alleged ROSS bypassed subscription restrictions to rush a competing product to market without investing the resources to build its own dataset. A federal district court sided with Reuters in 2025, rejecting the startup's fair use defense and ruling that ROSS infringed on copyrighted material.
ROSS appealed the decision to the Third Circuit, framing the case as a critical test for artificial intelligence development in the United States. Mark S. Davies, an attorney for ROSS, warned the panel about the broader implications. "Fair use ruling here brings into question the core technology of the AI revolution," Davies said.
Copyrightability of headnotes
The core legal debate centers on whether Westlaw headnotes qualify for copyright protection. Headnotes are short summaries of key legal issues that appear before judicial opinions, which themselves are public domain works. ROSS argued that these editorial additions lack the originality required for copyright and that copying them to teach a machine to answer legal questions constitutes quintessential fair use.
U.S. Circuit Judge Tamika Montgomery-Reeves questioned why the headnotes would not be considered sufficiently original. Davies countered that the startup only copied 25,000 headnotes out of a library of 28 million, amounting to 0.08 percent of the total. He emphasized that the data consists of legal context rather than creative works like poetry or fiction.
For legal professionals evaluating AI for Legal tools, the distinction between public law and proprietary editorial work is a frequent point of friction. Courts must balance the public's right to access the law with the commercial investments companies make to organize and summarize it.
Market harm and competitive intent
Reuters attorney Dale Cendali argued that the Supreme Court has already established that headnotes are copyrightable material. She drew a sharp line between the law itself and the work of people writing about the law.
Cendali told the court that ROSS caused direct market harm by offering a substitute product that relied on Westlaw's exclusivity. "They had the cases; their own experts admitted that they could have done their own analysis themselves and come up with questions and question pairs to do the same thing," Cendali said. "They didn't need to copy us, they just wanted to."
Cendali noted that Reuters launched its own AI-based ranking algorithms in 2010, positioning ROSS as a latecomer rather than an innovator. She argued that copying the data was a choice driven by commercial gain, not technical necessity.
Why this matters for legal professionals
The Third Circuit's ruling will directly affect how legal tech vendors build and train their models. If courts rule that summarizing public domain case law is copyrightable, AI developers may face higher barriers to entry, potentially limiting the AI Learning Path for Paralegals and other automated research tools to well-funded incumbents. Conversely, a fair use ruling could encourage broader experimentation with legal datasets, though it may force database providers to tighten their data access controls even further.
Your membership also unlocks:
