Thomson Reuters Wins Early Court Battle Over AI and Copyright in Legal Tech

On Tuesday, US District Court of Delaware judge Stephanos Bibas issued a partial summary judgment favoring Thomson Reuters in its copyright infringement lawsuit against the legal AI startup, Ross Intelligence. Filed in 2020, this case is pivotal as it addresses the legality of AI tools and their training methods, particularly the use of copyrighted data sourced without permission.

Similar lawsuits against OpenAI, Microsoft, and other AI giants are currently progressing through the courts, raising critical questions about whether AI tools can assert a “fair use” defense when utilizing copyrighted material.

Importantly, as Judge Bibas pointed out, this case pertains to “non-generative” AI, differentiating it from generative AI tools like large language models (LLMs). Ross ceased operations in 2021, labeling the lawsuit as “spurious” while admitting it could not secure adequate funding amid the ongoing legal dispute.

As detailed previously by Wired, Judge Bibas stated in his decision that “None of Ross’s possible defenses holds water” against the copyright infringement claims, ultimately dismissing Ross’s fair-use defense. The ruling heavily weighed on how Ross’s use of copyrighted materials impacted the market value of the original work by creating a direct competitor.

Thomson Reuters initiated the lawsuit due to Ross’s use of its Westlaw search engine, which indexes a substantial amount of non-copyrightable material, such as legal decisions, while also integrating its proprietary content. For example, Westlaw’s headnotes — summaries crafted by human editors — are a distinctive feature that enhances the value of Westlaw subscriptions for legal professionals.

In developing its legal research search engine, Ross converted the annotations and headnotes into numerical data reflecting relationships among legal terms for its AI training. Judge Bibas described how, after Thomson Reuters rejected Ross’s licensing request for Westlaw content, Ross approached another provider, LegalEase, to acquire 25,000 Bulk Memos crafted by attorneys using Westlaw headnotes for training data.

Ross CEO Andrew Arruda asserted that the Westlaw data served as “added noise” and claimed the tool was intended to “recognize and extract answers directly from the law using machine learning.” However, after analyzing the similarity between the 2,830 Bulk Memo questions, headnotes, and judicial opinions, the judge concluded that the evidence of actual copying was “so obvious that no reasonable jury could find otherwise.”