Brett Trout
In a significant decision for copyright law and legal technology, Judge Stephanos Bibas granted partial summary judgment in favor of Thomson Reuters, rejecting ROSS Intelligence’s fair use defense in a closely watched lawsuit over the use of Westlaw’s legal content. The ruling underscores the growing tension between traditional legal research providers and emerging AI-driven legal tools.
Background: The Battle Between Westlaw and ROSS
Thomson Reuters, the owner of Westlaw, sued ROSS Intelligence, an AI-powered legal research startup, alleging copyright infringement. At the heart of the dispute was ROSS’s use of Westlaw’s headnotes—editorial summaries of case law—to train its AI-driven legal search tool.
When ROSS initially sought a license for Westlaw’s content and was denied, it turned to a third-party legal research service, LegalEase, to create training data. LegalEase, in turn, instructed lawyers to develop legal questions using Westlaw headnotes as reference material. ROSS used this lawyer-generated data—comprising roughly 25,000 “Bulk Memos”—to train its AI. Thomson Reuters argued that this amounted to copying protected content, leading to the lawsuit.
Key Takeaways from the Ruling
- Copyright Validity Affirmed
Acknowledging that the text of judicial opinions themselves are not copyrightable, Judge Bibas ruled each of Westlaw’s headnotes and Key Number System are sufficiently original to be protected by copyright. He likened their creation to sculpting—removing unnecessary content to distill key legal principles, a process that involves sufficient creativity to warrant protection. He found that each headnote, even those that quoted judicial opinions verbatim, had original value as individual copyrightable works. - Direct Copyright Infringement Found
The court held that ROSS directly copied 2,243 Westlaw headnotes. By analyzing the substantial similarity between Westlaw’s headnotes and ROSS’s AI training data, Judge Bibas determined that ROSS had engaged in impermissible copying, not independent summarization of judicial opinions. - Fair Use Defense Rejected
ROSS’s fair use argument was a major focal point in this case. Judge Bibas ruled against ROSS on two of the four fair use factors:- Purpose and Character of Use: ROSS’s use was commercial and not transformative. While ROSS claimed it was using headnotes to train an AI, the judge ruled that its AI performed a similar function to Westlaw—assisting with legal research—rather than creating something new.
- Nature of the Work: This factor slightly favored ROSS, as Westlaw’s headnotes are not highly creative. However, the judge noted that this factor is rarely decisive in fair use cases.
- Amount and Substantiality of Use: ROSS did not incorporate Westlaw headnotes into its final product but still relied heavily on them for AI training. While this factor also favored ROSS, the judge concluded that the first and fourth factors were far more important.
- Market Impact: For this most important of all of the factors, the judge emphasized that ROSS’s AI directly competed with Westlaw, harming Thomson Reuters’s market and potential licensing opportunities.
- Broader Implications for AI and Copyright
The ruling reinforces the idea that AI companies cannot freely use copyrighted material for training without a license, even if the content does not appear in the final product. This could have significant implications for AI companies in legal tech and beyond, as courts increasingly scrutinize how copyrighted works are used in AI training.
What’s Next?
While ROSS may still appeal, this decision gives some current indication of how courts intend to examine the issue of copyright infringement in the AI training process. It also serves as a guide to direct AI developers as to how they should approach source training data to avoid claims of copyright infringement.
Recent Comments