Fair Use for Legal Tech
John Bowie
A US federal court’s landmark decision relating to fair use in AI technology in Thomson Reuters v. ROSS Intelligence offers some critical guidance for navigating AI and copyright law, signaling how courts may assess “fair use” in the age of machine learning.
Judge Stephanos Bibas ruled that ROSS infringed 2,443 Westlaw headnotes, rejecting its fair use defense because the startup’s AI-driven legal research tool commercially competed with Thomson Reuters without transforming the copyrighted material. While narrow, the ruling clarifies key boundaries for AI developers and content creators.
Why This Matters for AI Development
The court’s analysis of fair use’s four factors provides a roadmap. Bibas emphasized that ROSS’s use was non-transformative (Factor 1) and threatened Westlaw’s market (Factor 4), favoring Thomson Reuters.
But ROSS won on Factors 2 and 3: the headnotes’ factual nature and the absence of direct reproduction in outputs.
Chris Mammen of Womble Bond Dickinson said the split outcome underscores the need for nuanced arguments—for example, highlighting how training data is repurposed. “If you’re using creative works, expect stronger copyright claims,” he explains. Fact-heavy content? Fair use arguments gain traction.
Licensing Deals and Market Shifts
The decision could accelerate licensing agreements between AI firms and content providers. Alon Yamin of Copyleaks predicts a surge in paid partnerships, similar to OpenAI’s deals, creating new revenue streams for publishers.
Smaller startups, however, face higher stakes. ROSS’s closure amid litigation costs serves as a cautionary tale: robust compliance processes are non-negotiable. Yamin stresses transparency in training data sources, urging startups to “audit their pipelines or risk becoming the next casualty.”
The Bottom Line
This case isn’t just about headnotes—it’s also a blueprint for AI accountability. Companies must now prove their use of copyrighted material adds transformative value and doesn’t undercut the original market.
For lawyers advising clients, the message is clear that document the creativity in AI outputs, secure licenses where possible, and prepare to litigate the “transformative” threshold.
As Mammen put it, “You can’t copyright a block of marble, but carve something new—that’s where protection begins.”
