On Tuesday, US District Court of Delaware judge Stephanos Bibas issued a partial summary judgment in favor of Thomson Reuters in its copyright infringement lawsuit against Ross Intelligence, a legal AI startup. Filed in 2020, it’s one of the first cases that will deal with the legality of AI tools and how they are trained, often using copyrighted data scraped from somewhere else without license or permission.
Similar lawsuits against OpenAI, Microsoft, and other AI giants are currently winding their way through the courts, and they could come down to similar questions about whether or not the AI tools can claim a “fair use” defense of using copyrighted material.
In a statement given to The Verge by Thomson Reuters spokesperson Jeff McCoy, the company said:
We are pleased that the court granted summary judgment in our favor and concluded that Westlaw’s editorial content created and maintained by our attorney editors, is protected by copyright and cannot be used without our consent. The copying of our content was not “fair use.”
However, as the judge noted, this case involved “non-generative” AI, not a generative AI tool like an LLM. Ross shut down in 2021, calling the lawsuit “spurious” but saying it was unable to raise enough funding to keep going while caught up in a legal battle.
As reported previously by Wired, today Judge Bibas wrote in his decision, “None of Ross’s possible defenses holds water” against accusations of copyright infringement, and ultimately rejected Ross’s fair-use defense, relying heavily on the factor of how Ross’s use of copyrighted material affected the market for the original work’s value by building a direct competitor.
Thomson Reuters sued over Ross’s use of its Westlaw search engine. Westlaw indexes a good deal of material that is not copyrightable (like legal decisions) but also intersperses it with its own content. For instance, Westlaw headnotes — which are summaries of points of law written by human editors — are a signature feature meant to make the very expensive Westlaw subscription attractive to lawyers.
In building a legal research search engine, Ross turned the annotations and headnotes “into numerical data about the relationships among legal words to feed into its AI,” wrote Bibas. The ruling describes how, after Thomson Reuters rejected its attempt to license Westlaw’s content, Ross turned to another company, LegalEase, and purchased 25,000 Bulk Memos of questions and answers written by lawyers using the Westlaw headnotes that it used for training data.
Ross CEO Andrew Arruda claimed the Westlaw data was “added noise” and that its tool “aims to recognize and extract answers directly from the law using machine learning.” However, after having “compared how similar each of the 2,830 Bulk Memo questions, headnotes, and judicial opinions are, one by one,” the judge said the evidence of actual copying was “so obvious that no reasonable jury could find otherwise.”
