Westlaw vs. ROSS: The Copyright Battle Shaping AI

and not just for legal research

We have just seen the first major decision in the battles over using copyrighted material to train artificial intelligence models that could shape the future of AI-driven tools. The case involved two legal publishers, Thomson Reuters - the publishing giant behind the Westlaw database - and ROSS Intelligence - a startup using AI to analyse case law. A US federal judge ruled that ROSS copied thousands of Westlaw’s proprietary headnotes to train its AI and rejected ROSS’s defence that this was “fair use”

So, what happened? And what does this mean for the future of AI in legal tech? Let’s break it down.

⚖️ The Case: How an AI Startup Got Sued by Westlaw

📌 The Players:

• Thomson Reuters owns Westlaw, one of the largest legal research platforms in the world. It provides headnotes—short summaries of court decisions —that help lawyers find relevant cases.

• ROSS Intelligence built a legal AI tool to answer legal questions by analysing case law.

📌 The Problem:

ROSS wanted to license Westlaw’s data to train its AI model to assist in legal research. However, Westlaw refused. The reasons for the refusal were not revealed in the decision, but this refusal was probably because ROSS was building a competing product to Westlaw’s product. ROSS hired a third party (LegalEase) to create training materials using Westlaw’s headnotes.

Thomson Reuters found out—and sued.

💡 Key Legal Question boiled down to: Did ROSS infringe Westlaw’s copyright by using headnotes in training its own model? Or was this fair use of the summaries?

🏛️ The Court’s Ruling: This was a big win for Westlaw. The Judge ruled that ROSS infringed Westlaw’s copyrights.

✅ Westlaw Wins on Copyright Infringement

The court found 2,243 instances in which ROSS copied Westlaw’s headnotes through its AI training data. The court found that, even though judicial opinions are not protected by copyright law, Westlaw’s headnotes are “creative summaries” and were indeed protected by copyright.

❌ ROSS’s Defenses Rejected

ROSS tried several arguments, but the judge shot them all down: The argument that “we didn’t mean to infringe.” was not acceptable. Ross had access to Westlaw and used the database anyway. The judge confirmed that the ideas in the headnotes were copyrightable.” It was the way that the authors in Westlaw wrote and organized the headnotes that was protected.

Finally, and most significantly for other cases, the argument that “we were making fair use of the headnotes” was held not to be an excuse. This was not considered to be “fair use” of the copyrighted material.

🤔 Why ROSS’s Fair Use Argument Failed

Under U.S. copyright law, a court must look at four factors to determine fair use. Here’s how they played out in this case and are likely to be relevant in other cases:

1️⃣ Purpose & Character of Use (Favours Westlaw) ROSS used the headnotes commercially to train an AI that competes with Westlaw The court ruled this wasn’t transformative—ROSS used the headnotes for the same purpose as Westlaw (i.e. for simplifying legal research).

2️⃣ Nature of the Copyrighted Work (Favours Westlaw). Indeed, the headnotes were not “highly creative”, but they do at least meet the minimum originality requirement for copyright protection.

3️⃣ Amount & Substantiality (Favours ROSS). ROSS didn’t copy entire Westlaw documents, just portions of the headnotes, and the AI model did not directly output the headnotes—it simply used them for training.

4️⃣ Market Effect (Favours Westlaw—Most Important Factor). ROSS’s AI system was a competing product. Westlaw has a market for AI training data and so ROSS’s use could hurt that market.

🚨 Final Verdict: Since three factors (1,2 & 4) favoured Westlaw, the judge rejected ROSS’s fair use defence and ruled in Westlaw’s favour.

💡 Why This Case Matters for AI

This ruling isn’t just about Westlaw vs. ROSS—it’s a huge precedent for AI companies using copyrighted data. The judge has made it harder for AI companies to claim fair use when training models on proprietary databases and this will impact AI companies scraping content for training in the United States (such as OpenAI, Google, etc.). It will also have an international impact since it is relevant also to non-US companies wishing to sell their services into the US.

🔮 What’s Next?

1️⃣ ROSS may appeal—but with two fair use factors strongly against them, an appeal could be tough.

2️⃣ AI startups will tread carefully—we can expect to see more copyright licensing deals before AI companies use any databases.

3️⃣ Big publishers (LexisNexis, Bloomberg, SpringerNature, etc.) will be watching—this ruling gives them a stronger case to block AI training on their content.

🚀 Bottom Line: If you’re building AI using copyrighted material, you will need to pay for access to training data—or risk a lawsuit.

💬 What Do You Think?

🔥 Is this ruling good or bad for legal tech?

🔍 Should AI companies be able to train on legal headnotes without a license?