• by Wendy Davis  @wendyndavis, February 12, 2025

In what appears to be a first, a federal judge has found that an artificial intelligence company infringed copyright by training its service on material it didn't own.

The ruling, issued Tuesday by Judge Stephanos Bibas in Delaware federal court, comes in a battle between two companies that sell legal research tools -- Thomson Reuters' Westlaw and the competitor Ross Intelligence.

Westlaw publishes court opinions, and also publishes “headnotes” to those opinions. Those headnotes provide short summaries of the key points in the opinion. Ross trained its rival legal research platform on those Westlaw headnotes, according to the court ruling.

Tuesday's decision comes in a dispute dating to 2020, when Thomson Reuters alleged that Ross “surreptitiously” obtained Westlaw's headnotes from a licensee, and then used them “to rush out a competing product without having to spend the resources, creative energy, and time to create it itself.”

“The net result,” Thomson Reuters wrote in the complaint, “is that plaintiffs are now being put in the unfair position of having to compete with a product that they unknowingly helped create.”

Ross argued that it should prevail for numerous reasons, including that it made fair use of the material.

Bibas rejected that argument, essentially holding that Ross's use of the material didn't qualify for fair use because the company used Westlaw's headnotes for the same purpose as Thomson Reuters, and because Ross intended to develop a product that would substitute for Westlaw's in the market.

“Ross’s use is not transformative because it does not have a 'further purpose or different character' from Thomson Reuters’s,” he wrote.

“Ross’s use is commercial. Ross admits as much,” Bibas added.

Bibas's decision hinges on the facts of the dispute, but the ruling could have implications for lawsuits by other copyright owners over the use of their material to train artificial intelligence, according to Jeremy Goldman, a partner in the law firm Frankfurt Kurnit Klein & Selz.

Among the most high profile of those lawsuits are ones brought by newspaper publishers and book authors against OpenAI, which allegedly ingested copyrighted material to train its large language model, ChatGPT.

Those cases, however, differ in at least one important respect from the Westlaw-Ross dispute, Goldman tells MediaPost.

He says companies like OpenAI argue that they didn't copy material to compete with authors or newspapers, but to create a large language model, and that doing so requires them to analyze as much data as possible in order to understand language.

“Their position is, and has been, that they need to ingest all the data they can get their hands on in order for the technology to work,” Goldman says.

In other words, he says, OpenAI and other generative artificial intelligence companies will argue that they have a “big purpose” and therefore “need a big corpus of data.”