By Sam Savanich and Jeff Whittle

Technology often outpaces the law, but a new copyright infringement decision in the U.S. District Court for the District of Delaware shows that the courts are catching up in regulating artificial intelligence (AI), particularly around the use of certain data for training AI tools.

In Thompson Reuters v. Ross Intelligence, the court ruled that Ross overstepped its bounds by using short summaries similar to Thomson Reuters’ headnotes to train its AI legal research tool. The court found in favor of Thomson Reuters on two of four factors of the “fair use” defense to copyright infringement and ultimately ruled in favor of Thomson Reuters in regard to copyright infringement, at least for a portion of the total headnotes.

Ross used data similar to headnotes from Thomson Reuters’ Westlaw Keynote System, obtained via “Bulk Memos” from LegalEase, for training their AI tool. It is important to reiterate at this point that the “Bulk Memos” that copied the Westlaw headnotes were not presented to the end-user of Ross’ legal search tool, rather, the “Bulk Memos” were only used for training. After Thomson Reuters sued Ross for copyright infringement, Ross claimed fair use, as well as several other defenses, which will be explained below, along with the court’s interpretation of these defenses and other issues. Interestingly, Judge Stephanos Bibas ruled in 2023 that the AI tool did not infringe upon Thomson Reuters’ copyrights. In this latest ruling, the Judge determined, for a portion of the headnotes, that Ross did not meet their burden for fair use and thus infringed on Thomson Reuters’ copyrights. The Judge wrote in the February 11, 2025 decision, “Wisdom does not always find me, so I try to embrace it when it does––even if it comes late, as it did here.”

Direct Copyright Infringement

Thomson Reuters claimed, as noted, that Ross directly infringed its copyrights. Thomson Reuters ownership of copyright registrations regarding the headnotes in question is not in dispute. However, the court did analyze whether the headnotes were protectable or could be covered by a copyright. Reconsidering the previous opinion that the originality of the headnotes was a question for a jury, the Judge described the headnotes, at least over two thousand of the many headnotes in question, as including enough originality to meet the threshold for copyright validity. The Judge analogized Thomson Reuters’ role in creating the headnotes to that of a sculptor, writing, “A block of raw marble, like a judicial opinion, is not copyrightable. Yet a sculptor creates a sculpture by choosing what to cut away and what to leave in place. That sculpture is copyrightable.” Thus, the court granted summary judgment for Thomson Reuters on whether the headnotes were original enough to prevent rebuttal of any presumption of validity.

Copying of Original Elements

Ross claimed that, since the headnotes are less protectable expressions of work, the questions in the “Bulk Memos” must be virtually identical to the original headnotes. The court, in applying that standard, found this to be the case. The court stated that the questions in the “Bulk Memos”, more specifically 2,243 of 2,830 questions, were so substantially similar, that no reasonable jury could find otherwise. As shown by at least one example in the opinion, the only differences between the questions and the headnotes are the insertion of one or two words and a question mark.

Ross’ Defense

The court briefly analyzed four different defenses presented by Ross and deemed each to fail. First, innocent infringement, which only limits damages, failed due to the copyright notice included with the headnotes. Second, Ross alleged that Thomson Reuters “weaponized” their copyright against the public interest, however, the court found that Ross failed to show how Thomson Reuters misused its copyright to stifle competition. Third, the court deemed the merger defense inapt, stating that there are many ways to “express points of law from judicial opinions.” Finally, the court deemed the scenes à faire defense a failure since “nothing about a judicial opinion requires it to be slimmed down to Thomson Reuters’s headnotes”.

Fair Use Factors

In analyzing four of the factors of the fair use doctrine, the court ruled in favor of Thomson Reuters, noting that the first and fourth factors analyzed were given the most weight. First, the court stated that Ross’ purpose was commercial as Ross was producing some end product that utilized the copywritten material. Further, the court ruled that Ross’ use was not transformative as the use of the headnotes was to create a competing legal research tool. The court also noted that, even if there was no bad faith in the use of the “Bulk Memos” and since the use was commercial and non-transformative, the lack of bad faith would not “move the needle” or outweigh. Second, the court noted that the headnotes included limited creativity, conceding this point to Ross. Third, the court noted that Ross did not copy many of the large number of headnotes, thus conceding this point to Ross as well. Finally, the court considered the impact of Ross’ AI tool on the market. Ross’ AI tool directly competed with Thomson Reuters’ services, thus heavily impacting the market for Thomson Reuters. Lending the most weight to the last factor analyzed, the judge, as noted, ruled against Ross and granted summary judgment to Thomson Reuters on direct copyright infringement for at least some of the headnotes.

Take Aways

For those who train models, one should ensure that the data utilized for training are either obtained via a license and/or, at the least, not under copyright (e.g., in the public domain or purely functional). For those who own copyrighted data, some research as to what AI tools are available is recommended. In other words, based on a type of AI tool, one can assume what type of data may have been used to train that tool. Based on the facts in this case, Thomson Reuters knew that Ross was at least attempting to use data similar to their headnotes, as Ross tried to obtain a license to those headnotes via a license agreement and Thomson Reuters refused, since Ross was a direct competitor. If data that are being used to train an AI tool are suspected to be potentially similar to one’s data, then research and investigation should be performed. Such an investigation may include querying the AI tool and reviewing user manuals or documentation related to the AI tool, among other steps. Finally, ensuring that data are covered by copyright at least enables one to seek protection of their data, should misuse be suspected.

This article is for informational purposes and does not constitute legal advice.
The views expressed do not necessarily reflect the views of LES or Womble Bond Dickinson.

Share this article: