As Artificial Intelligence (‘AI’) continues to reshape industries, legal disputes over intellectual property (‘IP’) — particularly copyright — are becoming more frequent. Recently, the US District Court of Delaware issued a significant ruling on whether AI (specifically non-generative AI) can be trained on copyrighted legal materials without permission.
This case has ignited debate in both the IP and AI communities, but one big question remains: How will this ruling impact generative AI in the future?
Let’s explore.
What Happened?
Judge Stephanos Bibas summarised it best in his ruling:
Ross Makes a Legal AI Tool and Westlaw’s Owner Sues.
This case centres on Ross Intelligence, a legal research startup that aimed to compete with Westlaw, which is owned by Thomson Reuters.
For context, Westlaw is a leading legal research platform that provides paid access to, amongst others, case law, statutes and regulations. It also contains value added editorial content, such as headnotes, which summarise key legal principles and case holdings. Thomson Reuters owns copyrights of these value added editorial content.
Ross, needing a database of legal questions and answers to train its AI, requested a license from Westlaw, but was denied (for obvious reasons).
Instead, Ross allegedly obtained legal data from Westlaw via a third-party contractor, LegalEase, in the form of ‘Bulk Memos’. These memos were created by lawyers following LegalEase’s guidelines, which referenced Westlaw headnotes. Ross then used these memos to train its AI search tool.
When Thomson Reuters discovered this, they initiated an action against Ross for copyright infringement.
Key Issues
Are legal headnotes copyrightable?
The court ruled that Westlaw’s headnotes are copyright-protected as a compilation, even though the underlying case law itself is found in the public domain.
Does AI training qualify as fair use?
The court rejected Ross’ fair use defence, stating that its AI was built to compete with Westlaw, making its use commercial and non-transformative. Judge Bibas noted that Ross’ use did not serve a ‘further purpose’ or have a ‘different character’ from Westlaw’s own use of headnotes.
What does this mean for AI development?
The ruling sets a precedent, ie using proprietary legal data for non-generative AI training without permission could lead to copyright infringement.
Interestingly, the Judge explicitly noted:
Because the AI landscape is changing rapidly, I note for readers that only non-generative AI is before me today
This raises the question: How will courts treat generative AI that synthesises information rather than retrieving it?
Why This Matters
- AI companies must be cautious when sourcing training data, as copyrighted content is not freely available for use.
- Legaltech startups need clear IP strategies to avoid potential lawsuits that could threaten their survival.
- The ruling may impact AI-driven platforms in healthcare, finance, education, and other industries that rely on proprietary datasets.
- The ruling may impact AI-driven platforms in healthcare, finance, education, and other industries that rely on proprietary datasets.
Key Takeaway: The intersection of AI, copyright, and fair use (or ‘fair dealing’ in Malaysia) is still evolving. Businesses leveraging AI must navigate IP risks carefully or face potential legal challenges.
