Share This Article
We explore the landmark Delaware Court decision in the Thomson Reuters v. Ross Intelligence which analyzes the implications for AI training data practices under the US fair use copyright exception, and address how the outcome might differ under the EU’s Text and Data Mining (TDM) exceptions.
You can explore the topic in the latest episode of our podcast Diritto al Digitale, featuring myself Giulio Coraggio and our intellectual property DLA Piper colleague Valentina Mazza available HERE and below
Also you can know more on the matter in the article below
The Background
At the heart of this dispute was Ross Intelligence, a company developing AI-powered legal research tools, and Thomson Reuters, owner of Westlaw, a widely-used legal research platform. After Reuters refused Ross Intelligence’s request for a content license, Ross turned to LegalEase, obtaining training data known as “Bulk Memos”—documents derived from Westlaw’s copyrighted “headnotes.” When Thomson Reuters discovered this indirect use, it filed suit, alleging copyright infringement.
The Court’s Decision: Why it Matters
Applying the established four-factor fair use test under U.S. copyright law, the Delaware court concluded:
- Commercial Purpose and Non-transformative Use: Ross’s goal was commercial, aiming directly to compete with Westlaw by using its proprietary content. Simply converting text into numerical data for AI training did not transform the material sufficiently to justify fair use.
- Nature and Extent of Copied Works: While Westlaw’s headnotes were protected content, their nature—summaries of legal decisions—meant they were less creative, tipping this factor slightly in Ross’s favor.
- Market Impact: Critically, Ross’s AI product was deemed a direct market substitute for Westlaw, threatening its business model and future revenue opportunities in AI-driven legal research.
Balancing these considerations, the court decisively ruled against Ross Intelligence, setting a pivotal precedent in U.S. copyright law.
Broader Implications for AI Development
This decision signals tougher scrutiny ahead for AI companies relying on copyrighted data. While not directly about generative AI (like ChatGPT), the ruling tightens fair use parameters, increasing the potential legal risk for developers. Essentially, companies must now reconsider how they source and handle training data, especially if the AI product competes directly with the original content provider.
Could the EU Approach Differ under the Text and Data Mining Copyright Exception?
In the EU, where copyright law explicitly addresses Text and Data Mining (TDM), the outcome might differ, but challenges remain:
- Lawful Access Requirement: EU law requires clear lawful access to content for TDM. Given Reuters’ denial of access, Ross might face similar difficulties.
- Rightsholder Opt-out: The EU’s Copyright Directive allows rights holders explicit opt-out mechanisms. If Westlaw had properly exercised this right, Ross would again face significant hurdles.
- Storage Limitations: The directive restricts data retention to the duration of mining activity. Any long-term storage or reuse by Ross could lead to legal conflicts.
Thus, despite distinct frameworks, an EU court could similarly lean against fair use.
Future Directions: Bridging US and EU Approaches?
Interestingly, this U.S. decision reflects a shift toward the EU’s stricter stance. Traditionally more permissive due to fair use doctrine flexibility, the U.S. system now seems to embrace tighter controls protecting rights holders’ markets—a cornerstone of the EU model.
Looking forward, the intersection between AI innovation and copyright protection poses challenges. EU judges and lawmakers will likely need to clarify how intermediate uses—such as indirectly obtained training data—fit into existing frameworks. Additionally, evolving regulations like the AI Act promise more stringent transparency requirements, potentially reshaping how AI developers handle copyrighted materials.
In short, while this Delaware decision might currently set the stage for heightened caution, it could also encourage clearer rules and safer practices, influencing global legal frameworks around AI training and fair use.
You can read more articles on the topic HERE.
