Thomson Reuters Will Head to Trial in AI Model Copyright Battle

Thomson Reuters Enterprise Centre GmbH will go to trial next year against the legal tech startup Ross Intelligence Inc. in a case examining whether copyright law can limit artificial intelligence models from consuming copyrighted data without permission.

Reuters, which hosts and publishes the popular legal research platform Westlaw, failed to convince a federal judge in Delaware that Ross had infringed its copyrights by hiring another company to copy Westlaw “headnotes” and Key Number System to be fed into a machine learning model.

Judge Stephanos Bibas on Monday also declined to rule in favor of Ross, finding that only a jury can evaluate all four statutory factors under copyright law’s fair use defense. A trial is tentatively set for May 2024. Bibas, a circuit court judge, sat on the US District Court for the District of Delaware by designation.

Reuters said in a statement that “this case continues to be about ROSS’ theft of Thomson Reuters proprietary commentary, analysis, and organizational system” and that it looks “forward to presenting the evidence to a jury.”

Ross didn’t immediately return a request for comment.

The case is among the first to test whether copyright owners can prevent AI companies from sweeping up copyrighted works for the purpose of training the machine learning models behind powerful generative AI programs. Other companies including OpenAI Inc., Meta Platforms Inc., and Midjourney Inc. have been hit with a flurry of copyright lawsuits from authors and visual artists this year.

Ross sought to create an AI-based legal search engine that would allow users to enter questions and receive direct quotations from judicial opinions. The company first attempted to get a license to use Westlaw headnotes—short summaries of legal points that appear in opinions—but Reuters declined.

Ross then contracted a third-party to create a set of 25,000 question-and-answer sets for Ross’ machine learning model to train on. Reuters sued for copyright infringement in 2020, alleging that the questions were direct copies of Westlaw headnotes with a question mark at the end.

Ross described its use of the headnotes as transformative under the first fair use factor: it converted the plain language into numerical data, which it then fed into a machine learning algorithm “to teach the artificial intelligence about legal language.”

“It was transformative intermediate copying if Ross’s AI only studied the language patterns in the headnotes to learn how to produce judicial opinion quotes,” Bibas said in his order mostly denying both parties’ motions for summary judgment. “But if Thomson Reuters is right that Ross used the untransformed text of headnotes to get its AI to replicate and reproduce the creative drafting done by Westlaw’s attorney-editors,” then Ross can’t point to transformativeness.

That is a factual question for a jury to determine, Bibas said.

Bibas also said that although Ross admitted to copying some portions of the headnotes, factual questions remain about the copyrightability of the headnotes and the similarity between the headnotes and questions.

Bloomberg Law competes with Thomson Reuters in providing legal news, analysis, and workflow tools to the legal community.

Potter Anderson & Corroon LLP and Crowell & Moring LLP represent Ross. Kirkland & Ellis LLP and Morris Nichols Arsht & Tunnell LLP represent Reuters.

The case is Thomson Reuters Enter. Ctr. GmbH v. Ross Intel. Inc., D. Del., No. 1:20-cv-00613, 9/25/23.