Thomson Reuters sued Ross Intelligence for scraping Westlaw headnotes to train its AI legal research tool. The court granted summary judgment to Thomson Reuters, making this the only federal ruling to date holding that AI training on copyrighted material is not fair use.
Holding
Judge Stephanos Bibas granted summary judgment for Thomson Reuters, finding that Ross Intelligence's use of Westlaw headnotes to train a competing AI legal research product was not fair use. The court emphasized the commercial and substitutive nature of the use — Ross built a product that directly competed with the source material.
Arguments For / Positive Implications
- First and only federal ruling squarely holding that AI training can constitute copyright infringement
- Provides a counterweight to the emerging pro-fair-use consensus in AI training cases
- Protects investment in curated legal databases and editorial work product
- Draws an important distinction between training on raw works vs. curated editorial content
Arguments Against / Concerns
- May chill AI development in legal tech specifically, where training data is inherently legal content
- The ruling's reasoning is narrow — focused on the direct competitive relationship — limiting its broader applicability
- Does not address whether training on the underlying case law (vs. Westlaw's editorial additions) would be fair use
- Ross Intelligence shut down before trial, so the case was never fully litigated on the merits
Our Takes
This is the outlier everyone needs to read carefully. While other courts have found AI training to be fair use, Judge Bibas said no — because Ross was building a direct competitor using Thomson Reuters' editorial work product. The key distinction is that Westlaw headnotes are curated creative expression, not raw public domain material. This case doesn't kill AI training; it says you can't photocopy your competitor's homework.Lawra (The Moderate)
Finally, a court that sees through the 'transformative use' smoke screen. Ross Intelligence took Thomson Reuters' proprietary headnotes — the product of human editorial judgment — and fed them into a machine designed to replace Westlaw. That's not transformation; that's theft with extra steps. Every AI company building on scraped proprietary content should be worried.Lawrena (The Skeptic)
Let's keep perspective: this case is the exception, not the rule. Ross Intelligence made the rookie mistake of training on a direct competitor's proprietary editorial content to build a competing product. That's commercially substitutive on every level. The ruling says nothing about training on raw case law, published opinions, or works where the AI output serves a fundamentally different purpose. Smart AI companies will route around this easily.Lawrelai (The Enthusiast)
There's an important distinction this case highlights: case law itself is public domain — it belongs to everyone. But Westlaw's editorial headnotes are curated creative work. Ross's mistake wasn't using AI to make legal research better; it was building a competing product by copying another company's proprietary editorial output. The takeaway for legal tech innovators: train on the raw public record, add your own value layer, and compete on innovation — not on copying your competitor's homework. The market works when everyone plays by the same rules.Carlos Miranda Levy (The Curator)
Why This Case Matters
Thomson Reuters v. Ross Intelligence stands alone in the federal courts. While every other major ruling on AI training has found fair use (or at least leaned toward it), Judge Bibas went the other direction. This makes the case essential reading for anyone trying to understand where the legal lines around AI training actually are — because there is clearly more than one line.
What Happened
Ross Intelligence, an AI-powered legal research startup, sought to build a competitor to Westlaw. To train its natural language search engine, Ross used a third party to scrape thousands of Westlaw headnotes — the short editorial summaries that Thomson Reuters’ attorney-editors write to describe the legal principles in each case. Thomson Reuters sued for copyright infringement in 2020.
Ross Intelligence argued that its use was transformative fair use — it was building a fundamentally new product (an AI search tool) rather than reproducing the headnotes themselves. The court disagreed.
The Fair Use Analysis
Judge Bibas applied the four-factor fair use test and found each factor weighed against Ross:
-
Purpose and character: Ross used the headnotes for essentially the same purpose Thomson Reuters created them — to help legal researchers find relevant case law. The AI overlay did not make the use sufficiently transformative.
-
Nature of the work: Westlaw headnotes involve significant editorial judgment and creative expression, making them more deserving of copyright protection than raw factual compilations.
-
Amount used: Ross scraped thousands of headnotes — a substantial portion of Thomson Reuters’ editorial database.
-
Market effect: Ross was building a product that directly competed with Westlaw, the very product the headnotes were created to support. This was the strongest factor against fair use.
The Broader Impact
Thomson Reuters v. Ross Intelligence is frequently cited alongside Bartz v. Anthropic and NYT v. OpenAI as the three cases that will define AI copyright law. But unlike Bartz (which drew a line at piracy while endorsing fair use for purchased works) and NYT (which remains ongoing), this case flatly rejected the fair use defense for AI training.
The practical takeaway for AI companies: training on a competitor’s proprietary editorial content to build a competing product is the highest-risk scenario. The further your use case moves from that pattern — toward non-competing applications, raw public domain sources, or licensed data — the safer you are.
Sources
- Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc., No. 1:20-cv-00613-SB (D. Del. Feb. 11, 2025) (2025-02-11)
- Thomson Reuters Wins Landmark AI Copyright Case Against Ross Intelligence — Reuters (2025-02-11)
- The First AI Training Copyright Loss: What Thomson Reuters v. Ross Means — Columbia Journal of Law & the Arts (2025-03)
Explore Legal Frameworks
Cases don't happen in a vacuum. Explore the regulatory frameworks shaping AI law around the world — from the EU AI Act to emerging legislation in Latin America.
Ready for structured learning? Explore the Learning Program →
Comments
Loading comments...