Thomson Reuters lawsuit win not telling of other AI fair use cases
A judge rules in favor of the media company. However, the decision will likely not significantly affect generative AI lawsuits still waiting for a court ruling.
Thomson Reuters' legal victory over AI startup Ross Intelligence in an AI copyright infringement case shows that the fair use argument by AI vendors in copyright cases is complicated and varies from case to case.
While the decision is the first major ruling in an AI copyright case, according to some legal observers it will likely have little impact on other copyright infringement cases because this one involved traditional, not generative AI.
A federal judge on Feb. 11 ruled in favor of the news and media company in its copyright lawsuit against legal AI startup Ross Intelligence, which is now out of business. In the suit, Thomson Reuters argued that the startup used the publisher's legal research platform Westlaw to train Ross Intelligence's AI system. Ross Intelligence argued that its training was fair use -- the principle that permits unauthorized use of copyrighted content in certain circumstances -- an argument other AI vendors have used in copyright cases.
Tuesday's ruling is a revision of a judgment by U.S. Circuit Judge Stephanos Bibas in September on Thomson Reuters copyright infringement claims filed in 2020, in which the judge said the case should go to a jury.
In this week's ruling, Bibas disagreed with Ross Intelligence's fair use claim, but pointed out that the ruling had to do with the use of nongenerative AI rather than generative AI. He noted that Ross Intelligence copied part of Thomson Reuters' legal research platform Westlaw's headnotes -- a summary of key legal claims in a case -- to train its AI system. The judge also said Ross' use of Thomson Reuters' headnotes did not produce a different type of content than the original, and the AI startup used it to create a competing product.
Generative vs. nongenerative
The Thomson Reuters case is one of many AI lawsuits content creators have filed since the popularization of generative AI starting with OpenAI's introduction of ChatGPT in 2022. In a major copyright infringement lawsuit, The New York Times is suing OpenAI and Microsoft. Generative AI vendors Stability AI and Midjourney are also facing class action lawsuits for data infringement.
It's by no means definitive for a number of the other cases that we're watching closely.
Aaron RubinPartner, Gunderson Dettmer
So, it is tempting to use the ruling in the Thomson Reuters case as an example of a possible outcome in other lawsuits. However, that would be wrong, said Aaron Rubin, a partner at law firm Gunderson Dettmer.
"This case has pointed in the direction of copyright infringement, but reading the opinion, it's clear it's by no means definitive for a number of the other cases that we're watching closely," Rubin said.
One reason is that the Thomson Reuters case concerns the nongenerative use of the media company's products. While OpenAI and other AI vendors argue that they used copyrighted materials from publishers in a way that makes the generated content different from the original, the judge argued that this wasn't so with Ross because the former company did not create new content with Thomson Reuters' headnotes.
Transformative or not
"Maybe, if this were a generative AI technology at issue, the argument as to whether what's being created here using that underlying work, whether it's transformative or not, would be much more compelling than what Ross has to offer here," Rubin said.
The judge's emphasis on generative versus nongenerative technology also indicates that this case will likely not significantly affect generative AI copyright infringement cases, said Vincent Allen, partner at intellectual property law firm Carstens, Allen & Gourley.
"One thing it does show you is that fair use defense is not a simple, straightforward defense," Allen said. He added that a significant factor in the defense is whether copying the copyrighted material would lead to displacement of the original material. In some other lawsuits, vendors argue that the content created using copyrighted material often does not compete with the original work because new ideas were generated.
"The challenge for developers developing these solutions is making sure that they do a proper analysis of the fair use defense if that's what they're going to rely on," Allen continued. "It's probably a safer practice to try to license the data that you're going to use to train your model."
Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence software and systems.
