A federal court ruling in Thomson Reuters v. Ross Intelligence has set a significant precedent for AI companies. The Delaware District Court found that Ross’s AI system infringed copyright because it produced near-identical versions of Westlaw’s proprietary legal headnotes.
This ruling is a crucial reminder for product and tech teams integrating AI into their platforms to evaluate the legal foundation of their AI models. Choosing models trained on unlicensed data could introduce significant legal and business risks, affecting product viability and long-term scalability.
🚨 AI training ≠ fair use by default – AI developers often argue that AI training involves only incidental copying, meaning the copyrighted material is temporarily processed but not stored or reproduced in a way that directly competes with the original work. However, the court ruled that intermediate copying copyrighted text for AI training can still be an infringement.
🚨 Market competition matters – If an AI system performs the same function as the copyrighted work, it may be seen as a direct competitor.
🚨 Market harm is a decisive factor – The court ruled against fair use primarily because Ross’s AI competed with Westlaw in the legal research market. While it acknowledged a potential licensing market for AI training data, the ruling didn’t depend on it—Ross’s use infringed because it created a substitute product.
📌 Near-Identical Reproductions – The court found that Ross’s AI-generated outputs are substantially similar to Westlaw’s copyrighted headnotes.
📌 Direct Commercial Competition—The AI tool was built to replace Westlaw’s legal research service, making it more challenging to argue fair use.
📌 Market Harm & Licensing – The court emphasized that Ross’s AI competed directly in the same market as Westlaw, making the licensing discussion secondary. While the court acknowledged the potential for a licensing market, it wasn’t necessary to establish infringement in this case.
Unlike Ross’s system, Generative AI models don’t always produce near-verbatim copies of their training data. Instead, they generate new content based on patterns learned from large datasets.
It depends. Courts will likely consider two key factors:
1️⃣ How different is the AI-generated output from its training data?
2️⃣ Does the AI system replace or erode demand for the original content?
While Ross does not settle the fair use debate for GenAI, it signals a shift in how courts evaluate AI copyright cases: It’s not just about whether the AI creates something “new”—it’s also about whether it competes with the original work’s market.
This ruling raises a critical question for AI-driven product teams: Are you building on a legally sound foundation?
AI systems trained on unlicensed data introduce legal uncertainty and risk, which could lead to:
The Ross ruling underscores the importance of choosing AI models built exclusively on licensed data—ensuring compliance, reliability, and scalability.
At Bria, we saw these legal shifts coming and built our platform entirely on licensed data. This gives our customers a risk-free, future-proof foundation for AI innovation.
The bottom line? The Ross ruling is a wake-up call for AI developers: Technical transformation alone isn’t enough to claim fair use. AI companies must proactively address licensing, market overlap, and legal compliance to avoid costly legal battles.
Want to explore AI innovation built for the long run? Let’s talk. 🚀