The U.S. District Court for the District of Delaware delivered a watershed ruling in Thomson Reuters v. Ross Intelligence on February 11, 2025, providing clarity on an often-asked question: is the utilization of copyrighted subject matter covered by the fair use doctrine as an affirmative defense?
The decision, authored by Third Circuit Judge Stephanos Bibas sitting by designation, provides valuable insights into how courts are likely to analyze copyright disputes involving artificial intelligence systems employing large language models — particularly when the competitive commercial stakes are high.
Case Background: Competing Legal Research Tool Providers
Thomson Reuters, owner of the Westlaw legal research platform, sued Ross Intelligence for copyright infringement after Ross trained its AI-driven legal search to train on Westlaw’s copyrighted headnotes. These headnotes — attorney-authored summaries of legal determinations extracted from published judicial opinions — form the backbone of Westlaw’s proprietary Key Number System. This case does not turn on whether the system had trained on raw judicial decisions, which are not subject to copyright protection.
When Ross sought to license its competitor Westlaw’s content and was turned down, it contracted with LegalEase to create training data (“Bulk Memos”) drawn from Westlaw’s headnotes. Ross then used these memos to develop a competing AI tool that allowed users to search for related case law. The court found Ross’s Bulk Memos copied 2,243 copyright-protected headnotes verbatim or with only minor alterations, leading to partial summary judgment for Thomson Reuters on direct infringement.
The Fair Use Defense: A Four-Factor Breakdown
Judge Bibas’ analysis of Ross’s fair use defense under 17 U.S.C. § 107 offers a roadmap for evaluating AI training disputes, although, as is often the case, the facts underlying the decision matter. While all four statutory factors were weighed, the decision hinged on commercial purpose and market harm. The court’s infringement analysis logically built upon the Supreme Court’s emphasis in Warhol v. Goldsmith (2023) – determining that artist Andy Warhol’s Orange Prince, which was based on a copyrighted photograph, constituted fair use based on its distinct purpose being different from the original work. Also relevant in the analysis was Authors Guild v. Google (2021)– which concluded that Google’s wholesale digitizing of books for search indexing was deemed transformative.
1. The Purpose and Character of Use: Commercial Competition Undermines Claims of Transformation
In Thomson Reuters, the court rejected Ross’s argument that its use of headnotes was transformative because it converted them into numerical training data. While acknowledging that AI training can sometimes qualify as intermediate copying (as in software cases like Sega v. Accolade), Judge Bibas distinguished Ross’s actions:
- Non-transformative intent: Ross aimed to replicate Westlaw’s core functionality rather than create a new product. Its AI tool generated search results functionally equivalent to Westlaw’s headnote-driven outputs.
- Commercial substitution: The parties were direct competitors in the legal research market, with Ross explicitly seeking to undercut Westlaw’s pricing.
- Distinction from generative AI: The court noted that Ross’s tool was non-generative — it returned existing judicial opinions rather than synthesizing new content — weakening claims of transformative purpose.
Here, unlike Authors Guild, the lack of a “different character” or public-facing innovation proved fatal.
2. Nature of Copyrighted Work: Minimal Creativity Still Warrants Protection
Though Westlaw’s headnotes required editorial judgment, the court classified them as “factual compilations” with only modest creativity. This factor slightly favored Ross but carried limited weight compared to commercial impact.
3. Amount and Substantiality Used: Technical Copying vs. Public Exposure
While Ross ingested thousands of headnotes, the court found this factor neutral because the copied material never appeared in Ross’s public outputs. Citing Authors Guild, Judge Bibas emphasized that substantiality turns on what reaches end users — not the volume of material works developed internally.
4. Market Effect: The “Single Most Important” Factor
The ruling turned on evidence that Ross’s tool threatened Westlaw’s market position and emerging licensing opportunities. Key findings included:
- Direct competition: Ross’s product targeted the same customers as Westlaw with similar services offered by Westlaw.
- Licensing markets: Thomson Reuters had refused to license its content to Ross, signaling a protectable derivative market for AI training data.
- Public interest limitations: While society benefits from AI innovation, the court stressed that copyright incentivizes investments in tools like Westlaw. Ross could have developed its system without infringing by using public judicial opinions.
This analysis aligns with Harper & Row v. Nation Enterprises, where market harm outweighed public benefits.
Implications for AI Developers and Legal Practitioners
The Thompson Reuters decision clarifies several important points for AI-related copyright disputes:
- Commercial Intent Matters More Than Technical Process
Courts will scrutinize whether AI training serves a competitive commercial purpose. As Judge Bibas noted, “Transformative use arguments retain viability for generative AI,” but tools replicating existing markets face steep hurdles. - Licensing Markets Are Now a Key Battleground
The recognition of AI training as a potential licensing market undermines fair use claims when rights holders decline access. Developers must now weigh litigation risks against licensing costs (and be ready to move on if license terms cannot be reached between the parties). - Fact-Specific Analysis Limits Precedential Reach
The ruling explicitly distinguished non-generative AI systems like Ross’s from generative tools (e.g., ChatGPT). Courts may still find fair use in cases involving novel outputs rather than substitutes for the competitors’ copyrighted material. - Intermediate Copying Defenses Face Narrowing
While other cases (e.g., Sega v. Accolade and Google v. Oracle) which permitted copying content such as software code to achieve interoperability, the court here rejected this analogy for textual works, stating Ross could have legally accessed the underlying judicial opinions without any copyright violation.
Strategic Considerations for Legal Teams
For firms advising AI developers:
- Document transformative intent: Emphasize novel applications and outputs that don’t directly compete with training data sources.
- Explore licensing early: Early, strategic licensing may be a pragmatic and practical option to mitigate infringement claims, especially when dealing with curated content like Thomson’s case law databases.
- Monitor evolving standards: The Ross ruling leaves room for generative AI distinctions, while subsequent cases (e.g., NYT v. OpenAI ) will continue reshape the landscape.
For content creators:
- Explore derivative markets: Develop licensing strategies and frameworks for AI training to strengthen market-harm arguments.
- Monitor copying scope: Even internal ingestion of protected material may trigger liability if tied to commercial competition.
Conclusion: A New Era for AI and Copyright
The Thomson Reuters v. Ross Intelligence decision signals that courts will prioritize market realities over technological transformation in AI copyright disputes. While fair use likely still has a home in generative AI systems (at least for now), the ruling underscores a fundamental truth: when innovation directly threatens existing markets, copyright owners of underlying content hold the upper hand. As licensing regimes evolve and litigation proliferates, developers must navigate the realities of original content ownership with a facility for both technical and legal specifics.
What’s the takeaway for the AI industry? Transformation alone cannot override copyright’s core function in a business context, which is to protect investments in creative content from infringement, particularly by the copyright holder’s competitors.
To discuss further, please contact KJK Intellectual Property attorney, Ted Theofrastous (TCT@kjk.com).
