The Core Conflict: Substitution vs. Innovation

The 3rd Circuit Court of Appeals is currently reviewing a district court’s rejection of ROSS Intelligence’s fair-use defense regarding its use of Thomson Reuters’ copyrighted headnotes. The central tension lies in whether ROSS’s AI-driven legal research tool is a "transformative" technological advancement or a "classic case of substitution" that harms the market for Westlaw’s proprietary editorial content.

ROSS argues that its use was transformative because it utilized deep-learning techniques to teach a machine to understand the semantic relationships of legal language, rather than simply reproducing Westlaw’s content. Thomson Reuters contends that because both products serve the same purpose—helping users conduct legal research—the use is not transformative under the Supreme Court’s Warhol standard. The panel of judges focused heavily on the fourth fair-use factor (market harm), questioning whether ROSS’s product acts as a direct commercial substitute for Westlaw.

The Fair Use Balancing Act

During oral arguments, the panel probed several critical areas of copyright law:

  • Market Definition: The judges pressed both parties on the definition of the relevant market. ROSS argued that there is no distinct market for individual headnotes or for headnotes as AI training data. Conversely, Thomson Reuters identified three distinct areas of harm: direct competition with Westlaw, the loss of exclusive rights to train its own AI on its proprietary data, and the erosion of a potential licensing market for AI training.
  • The "Necessity" of Copying: The panel questioned whether ROSS’s copying was necessary. Unlike cases like Sega or Sony, where copying was required to achieve functional compatibility, the judges noted that ROSS could have theoretically created its own training materials from public-domain judicial opinions without relying on Westlaw’s editorial headnotes.
  • Editorial Originality: The court examined whether Westlaw’s headnotes possess sufficient originality to warrant copyright protection. While ROSS argued that headnotes are factual summaries designed to track judicial language, the judges suggested that the selection and condensation of opinion language reflect significant editorial judgment.
  • The Burden of Proof: ROSS emphasized that it copied only 0.08% of Westlaw’s 28 million headnotes. However, the judges reminded ROSS that it bears the burden of proving fair use, and Thomson Reuters argued that ROSS took the "heart" of the work—the question-and-answer pairs that provide Westlaw its competitive edge.

Though this case predates modern generative AI, the court’s ruling is expected to set a significant precedent for how AI developers train models on proprietary legal data. The panel’s focus on whether a product is "materially different" from existing tools suggests that the court is looking for a clear, factual distinction between "AI-enhanced" search and traditional legal research platforms. The court has not indicated a timeline for its decision.