Appeals Court Upholds Thomson Reuters Win in Ross AI Copyright Case

This case represents a critical juncture in the application of the fair use doctrine to generative AI and machine learning training datasets. Historically, copyright disputes involving databases often turned on the distinction between uncopyrightable facts and copyrightable expression or arrangement. In the context of AI, defendants have frequently relied on the 'transformative' prong of fair use, arguing that ingesting copyrighted works to create new functional models differs fundamentally from republishing the original content. Prior precedents in text mining and search indexing generally favored such uses when non-expressive. However, this dispute shifts focus toward market substitution and commercial competition as decisive factors. The pattern here suggests courts may distinguish between general-purpose AI training on open web data and targeted ingestion of proprietary, high-value vertical datasets where the output directly competes with the source. This aligns with evolving judicial skepticism toward uncompensated commercial exploitation of creative labor, moving away from broad technological exceptionalism toward traditional market-harm analysis.

On September 29, 2026, the U.S. Court of Appeals for the Third Circuit affirmed a lower court ruling in favor of Thomson Reuters in its copyright dispute against Ross Intelligence, according to multiple reports on Bluesky and legal news outlets. The appellate panel upheld the determination that Ross Intelligence’s use of copyrighted Westlaw headnotes to train its artificial intelligence legal research platform did not qualify as fair use under U.S. copyright law. This decision cements a significant precedent regarding the unauthorized commercial use of proprietary data for AI model training, specifically within the legal information sector.

The litigation traces back to May 6, 2020, when Thomson Reuters filed a complaint alleging that Ross Intelligence had copied thousands of headnotes from Westlaw to develop a competing AI-powered search engine. According to the initial filings and subsequent court summaries referenced in the provided sources, Thomson Reuters argued that this unauthorized replication undermined the economic incentives necessary to maintain comprehensive legal databases. Ross Intelligence, now defunct, had consistently contended that its use of the material was transformative fair use intended to promote innovation and improve access to justice through advanced legal technology.

In September 2023, a federal district court granted partial summary judgment to Thomson Reuters, rejecting Ross's fair use defense specifically regarding the replication of headnotes. The September 2026 appellate decision affirms that earlier ruling. Legal observers on Bluesky noted that while the outcome is clear, the full text of the Third Circuit's opinion remains sealed as of the reporting date. Brian L. Frye highlighted this procedural anomaly, questioning why the reasoning was not immediately public despite the landmark nature of the affirmation. Bloomberg Law confirmed that the court found Ross’s tool did not make fair use of the headnotes, describing it as a first-of-its-kind ruling.

Thomson Reuters successfully argued that the use of its content was not merely for research or analysis but for the direct creation of a commercial substitute. The appeals court’s rejection of the fair use argument signals that commercial replication of proprietary datasets for competing products faces high legal barriers. Conversely, critics and supporters of Ross Intelligence had framed the case as a test of whether copyright law would stifle AI innovation by locking up foundational legal knowledge. With the appellate win secured, Thomson Reuters has validated its legal strategy for protecting database assets against AI-driven competitors.

The immediate consequence is the solidification of liability risks for AI developers who utilize proprietary content without licensing. While the specific legal reasoning is currently inaccessible due to the seal, the operative holding establishes that the fair use doctrine does not automatically shield AI training processes involving copyrighted professional data. Industry stakeholders are now left to interpret the boundaries of permissible use based on the district court record and the appellate mandate until the full opinion is released.

AI startups and developers utilizing proprietary datasets for training face increased legal exposure and mandatory licensing costs, potentially stifling entry in specialized verticals like law, medicine, and finance. Incumbent data holders gain significant leverage to monetize content through litigation-backed licensing regimes. The magnitude extends beyond legal tech to any sector where AI models compete directly with source data providers. While Ross Intelligence is defunct, the precedent affects all current and future AI ventures relying on similar data ingestion strategies. Market dynamics may shift toward consolidated data partnerships, raising barriers for independent researchers and smaller firms unable to negotiate licenses or absorb litigation risks associated with fair use ambiguity.