On February 11, 2025, Judge Stephanos Bibas issued an opinion in Thomson Reuters Enterprise Centre GMBH v. Ross Intelligence Inc., civ. no. 1:20-cv-613, a dispute regarding copyright infringement allegations stemming from the use of copyrighted data from the Westlaw legal database used in the training of an AI search tool. Judge Bibas, sitting by designation in this matter from the Third Circuit, granted partial summary judgment to Thomson Reuters on direct copyright infringement and determined that Thomson Reuters' Westlaw key number system and more than 2,200 headnotes were impermissibly used to create Ross's competing product and are original enough to be protected by copyright.
After finding that the Westlaw system was copyrightable, Judge Bibas rejected Ross's fair-use defense. In the highly anticipated opinion, Judge Bibas found that two key factors of the "four factor fair use analysis" favored Thomson Reuters. Specifically, the purpose of Ross's use of headnotes from Thomson Reuters' Westlaw legal research service and its harm to the market for the originals both favor Thomson Reuters. Therefore, Ross's fair use defense failed, and using Westlaw's copyrighted headnotes to train Ross's AI search tool was found to be copyright infringement.
The case has taken many twists and turns. After engaging in discovery, the parties cross-moved for summary judgment initially in December 2022, with the court issuing an original order and opinion in September 2023.[1] At that time, the court largely denied the parties' motions as to Thomson Reuters' direct copyright claims and Ross's affirmative defenses, finding that it was the role of the jury to "sort through the evidence and tidy these factual messes."[2] Now, over a year later, and after receiving renewed motions for summary judgment, the court revisited various aspects of his original rulings, in summary holding:
- Thomson Reuters' alleged headnotes and Key Number System were original enough to satisfy the "extremely low" threshold required for copyrightability[3]
- Thomson Reuters established that Ross copied the original elements of a batch of 2,243 headnotes out of a batch of 2,830 headnotes[4]
- Thomson Reuters established that memoranda that Ross used to generate its AI search tool were substantially similar to 2,243 headnotes that Ross copied[5] and
- Ross failed to establish evidence supporting affirmative defenses of innocent infringement, copyright misuse, merger, scène à faire, or fair use[6]
To determine and evaluate actual copying, the court relied on expert opinions of the parties, and the court itself compared the materials used by Ross with Thomson Reuters' more than 2,000 headnotes individually.[7] In completing this analysis, the court stated that it granted summary judgment only as to the headnotes "for which actual copying is so obvious that no reasonable jury could find otherwise."[8] The court spent most of its opinion evaluating whether Ross's use of the alleged copyrighted content constituted "fair use," with the court ultimately concluding that the balance of the factors weighed against fair use.
Of note, in analyzing the purpose and character of Ross's use, the court held that Ross's use was commercial and not transformative. The court also declined to permit intermediate copying, and distinguished between prior computer-related cases where intermediate copying was permitted and the instant case.[9] The court reasoned that although the copyrighted material here was used to create an AI search tool, the copyrighted material was itself not a computer program, which courts have historically recognized differs from other literary works because computer programs serve a functional purpose.[10] Finding that the intermediate copying was not permissible in this case, the court further reasoned that the use of Thomson Reuters' alleged copyright material by Ross to generate its AI search tool for the likely purpose of competing with Thomson Reuters weighed against fair use.[11]
Judge Bibas previously denied Thomson Reuters' motion for summary judgment on infringement and against Ross's fair use defense, in part, because of the fourth factor analyzed under fair use, which deals with market impact. That previous decision was wrong, said the court after evaluating the renewed motions.
"Even taking all facts in favor of Ross, it meant to compete with Westlaw by developing a market substitute. And it does not matter whether Thomson Reuters has used the data to train its own legal search tools; the effect on a potential market for AI training data is enough," Judge Bibas elaborated in the opinion.
While the court's opinion as to copying and substantial similarity applies only to a subset of the alleged copyrights, it serves to narrow the issues at trial for a jury. Likewise, the court's reasoning may provide further guidance for non-party companies involved with creating AI search tools as to whether certain conduct will constitute infringement or will be permissible fair use. Given the rapidly changing nature of the AI landscape, the court was clear that "only non-generative AI" was evaluated in its opinion—whether fair use of the alleged copyrighted content would have been different if generative AI was involved remains to be seen.
If you have questions about using AI search tools or generative AI that incorporates elements from copyrighted works, contact Justin Pierce, Marcella Ballard, Maria Sinatra, or any professional on Venable's Intellectual Property team.
[1] Thomson Reuters Enterprise Centre GMBH v. Ross Intelligence Inc., civ. no. 1:20-cv-612 (SB), Dkts. 547 ("Original Summary Judgment Opinion").
[2] Original Summary judgment Opinion, at *34. The court further ruled on other claims and defenses raised by the parties that were not pursuant to the Copyright Act, which we do not discuss in this article for purposes of brevity.
[3] Second Summary Judgment Opinion, at *6.
[4] Id. at *12.
[5] Id. at *14.
[6] Id. at *14-16.
[7] Id. at *12. In evaluating copying, the court further explained that the parties agreed that Ross had access to Thomson Reuters' copyrighted content through a third party.
[8] Id. at *13.
[9] Id. at *18.
[10] Id. at *18-19.
[11] Id. at *22-23.