Third Circuit Rejects AI Fair Use Defense in Landmark Thomson Reuters Copyright Ruling

Inside the James A. Byrne United States Courthouse in Philadelphia on September 29, a panel of federal appellate judges handed down a sealed opinion that fundamentally alters the economics of artificial intelligence. The Third Circuit Court of Appeals ruled that an AI developer cannot claim fair use when it ingests a rival's proprietary data to build a competing product.[4][5]

The decision marks the first time a major US appellate court has squarely addressed the intersection of generative AI training and copyright infringement. It resolves a long-running dispute between legal publishing giant Thomson Reuters and the now-defunct AI startup ROSS Intelligence, setting a precedent that will shape the technology sector for years.[2][7]

Thomson Reuters, which operates the ubiquitous Westlaw legal research platform, initially sued ROSS in May 2020. The publisher alleged that the startup unlawfully copied thousands of Westlaw's proprietary headnotes—expert summaries of judicial opinions—to train a machine-learning model designed to answer natural-language legal queries.[6]

The Westlaw platform relies on a highly structured Key Number System, which employs human attorney-editors to read, summarize, and categorize every major judicial opinion in the United States. Thomson Reuters argued that ROSS bypassed the massive expense of this editorial process by simply downloading the finished summaries en masse.[6][7]

"Training an AI tool on Westlaw headnotes to compete with Westlaw is not fair use," attorneys for the law firm Mitchell Silberberg & Knupp wrote in their analysis of the appellate court's reasoning, highlighting the direct market threat posed by the software.[6]

The Commercial Substitute Test

At the heart of the litigation was the legal doctrine of fair use, which permits limited copying of protected works without permission under specific conditions. ROSS argued that its data ingestion was highly transformative because the AI did not reproduce the headnotes for end users, but merely analyzed their linguistic patterns to understand underlying legal concepts.[3][5]

The Third Circuit panel firmly rejected that defense, focusing heavily on the fourth factor of the statutory fair use test. That specific provision requires judges to examine the effect of the unauthorized copying on the potential market for, or value of, the original copyrighted work.[2]

Because ROSS deliberately targeted Westlaw's structured data to build a platform that lawyers would use instead of Westlaw, the court found the AI tool acted as a direct commercial substitute. The judges concluded that allowing such copying would destroy the financial incentive for publishers to create original reference materials.[2][7]

"Ease is not necessity," the court noted in its analysis, firmly dismantling the startup's argument that scraping the proprietary data was the only practical or financially viable way to train a highly capable legal artificial intelligence.[2]

The ruling draws a sharp distinction between a human reading a text to learn facts and a company systematically ingesting a structured database to replicate its commercial function. The court found that ROSS did not merely extract uncopyrightable ideas, but appropriated the specific editorial expression Thomson Reuters used to organize the law.[3][6]

Legal scholars note that the decision carefully avoids declaring all artificial intelligence training to be copyright infringement. Instead, it creates a highly fact-specific framework that looks at the relationship between the ingested training data and the final software product's market position.[7]

Ripple Effects Across the Tech Industry

The timing of the Third Circuit's decision arrives as the technology industry faces a massive wave of similar copyright lawsuits. Major developers are currently defending their data-scraping practices in federal courts against claims from prominent authors, news organizations, and software engineers.[4][7]

While the ROSS case involved a highly targeted dataset and a direct competitor, the appellate court's reasoning provides a powerful new weapon for plaintiffs in those broader disputes. It establishes binding precedent that intermediate copying for machine learning is not automatically shielded by the transformative use doctrine.[5]

Intellectual property attorneys expect the ruling to accelerate a shift toward licensed data agreements across the sector. Over the past 12 months, several leading artificial intelligence companies have signed multi-million-dollar contracts with traditional publishers to secure legal access to high-quality training archives.[6][7]

For companies building foundational models on public web data, the ruling offers a mix of warning and relief. Because the Third Circuit focused so heavily on ROSS building a direct substitute for Westlaw, AI developers argue the precedent should not apply to models that serve entirely different market functions than their training data.[7]

"The court gets merger wrong and does a disservice to fair use," argued the Authors Alliance, a group advocating for open access to information, warning that the ruling could allow large publishers to effectively monopolize access to basic legal facts.[3]

The Authors Alliance and other digital rights organizations had urged the court to protect the ability of researchers to computationally analyze texts. They maintain that the appellate decision dangerously conflates the uncopyrightable ideas contained in the headnotes with their copyrightable expression.[3]

Unsealing the Final Framework

The exact contours of the Third Circuit's 2026 ruling remain temporarily obscured by procedural redactions. The court issued the initial opinion under seal on September 29 to protect confidential business information submitted during the extensive discovery phase of the litigation.[4]

Attorneys for both Thomson Reuters and ROSS Intelligence have submitted proposed redactions to the appellate panel. IPWatchdog reports that the court is expected to release the public, unsealed version of the landmark decision in early October 2026.[4]

For ROSS Intelligence, the appellate ruling serves as a posthumous defeat. The startup ceased its commercial operations in December 2020, citing the overwhelming financial burden of defending against the Thomson Reuters litigation while trying to build its product.[2][5]

The broader technology sector now faces a fundamentally altered legal landscape. By prioritizing market impact over technological innovation, the Third Circuit has signaled that the era of unrestricted data scraping for commercial artificial intelligence development is facing strict judicial limits.[6][7]