Federal Court Decision Sheds New Light on AI Training Data and Copyright Law

A federal court’s recent decision in Thomson Reuters v. Ross Intelligence offers a rare moment of clarity at the murky intersection of artificial intelligence and intellectual property. More specifically, it provides guidance about the use of AI training data and copyright law.

In a February 2025 ruling, the District of Delaware granted partial summary judgment in favor of Thomson Reuters, finding that Ross Intelligence infringed the copyrights of 2,243 Westlaw headnotes by using them to train its own AI-powered legal research tool. Thomson Reuters owns Westlaw, an online legal research service and database.

The outcome is a significant development for companies building or deploying AI systems that rely on third-party content. The case centers on one of the most pressing legal questions in the AI era: can proprietary materials be used as training data without permission? This decision seems to make clear that the answer, in many instances, will be “no.”

The Case

Ross Intelligence set out to create an AI-driven legal search engine to rival Westlaw. When Thomson Reuters denied a license to use its database, Ross contracted with LegalEase to acquire around 25,000 “Bulk Memos”—legal questions and answers based, in part, on Westlaw’s headnotes. These memos were used to train Ross’s AI, which did not display headnotes to end users but relied on them to improve search functionality.

Thomson Reuters sued, alleging copyright infringement. After initially allowing the case to proceed toward trial, the court reversed course and granted partial summary judgment to Thomson Reuters. It found that Ross had copied 2,243 headnotes and that its use was not protected by fair use. The judge also rejected several other defenses, including innocent infringement and copyright misuse.

Key Takeaways

AI Use Does Not Immunize Against Copyright Law. This decision underscores that using materials for AI training—even at a background or “intermediate” stage—does not exempt companies from copyright law. Ross’s AI did not reproduce the headnotes in its final product, but that was not enough to avoid liability.

Fair Use Defense Faces Headwinds in Commercial AI Contexts. The court held that Ross’s use was commercial and non-transformative. Both factors weighed against a finding of fair use, especially since Ross aimed to develop a competing product. The opinion emphasized that fair use does not extend to uses that substitute the original in the market or undermine potential licensing revenue.

Editorial Content Is Protected. The court distinguished this case from those involving software code and interoperability. While prior decisions permitted intermediate copying of code for functional purposes, Westlaw’s headnotes were deemed editorial content, reflecting human judgment and creativity—making them eligible for copyright protection. Content of this nature carries a different level of legal risk when used in AI development.

Market Impact Is a Decisive Factor. Ross’s use of Westlaw content had the potential to harm both the existing market for legal research platforms and potential markets for AI training data. Courts will consider both actual and potential market harm when assessing fair use, particularly when the copied material enables a competing product.

Practical Implications

• Review Data Practices: Companies leveraging AI should assess the sources of their training data and ensure proper licensing is in place, particularly when using curated or proprietary content.

• Revisit Vendor Agreements: Contracts with AI vendors should include explicit provisions around data sourcing, intellectual property compliance, and indemnification for potential claims.

• Consider Licensing Strategies: For organizations that produce high-value content, this ruling highlights the potential to develop licensing models for AI training purposes—a growing and potentially lucrative market.

• Monitor Evolving Standards: Legal standards for AI are still developing, but this case illustrates how courts are likely to approach disputes where proprietary content is used to build competitive AI tools. Staying ahead of these developments is essential for risk management and strategic planning.

Looking Ahead

The decision in Thomson Reuters v. Ross marks a significant moment in the legal treatment of AI training practices. It affirms that traditional intellectual property principles remain fully in force, even as technology evolves rapidly. While this case focused on legal research tools, its implications extend across industries—from media and education to finance and healthcare—anywhere proprietary content meets AI development.

This ruling offers valuable guidance in an environment where innovation races ahead of regulation. Companies developing or deploying AI systems should view this decision as a clear signal: the data used to power AI matters, and how it is obtained can carry substantial legal risk.