Back

U.S. appeals court rules an AI startup broke copyright law by training its legal search tool on Westlaw's case summaries

An empty courtroom in the James A. Byrne U.S. Courthouse in Philadelphia, with a long judges' bench beneath the U.S. seal
Photo: Carol M. Highsmith, Library of Congress (public domain)

A federal appeals court has upheld a ruling that ROSS Intelligence broke copyright law when it used summaries written by Thomson Reuters' Westlaw editors to train an AI legal search tool built to compete with Westlaw. The court said ROSS could not rely on the legal exception that allows some copying without permission, because it copied to build a direct rival. The ruling is narrow: the court stressed that ROSS's tool only found passages in court opinions and could not write anything new, unlike chatbots.

A federal appeals court has ruled that an AI company broke copyright law when it copied a rival's writing to train its software. On Sept. 29, the U.S. Court of Appeals for the Third Circuit upheld a lower court's ruling for Thomson Reuters, which owns the legal research service Westlaw, against ROSS Intelligence, a startup that built an AI tool to compete with it. Many lawsuits ask whether AI companies may train their systems on copyrighted work without paying for it, and this ruling gives one appeals court's answer for one kind of AI.

What ROSS copied were headnotes, the short summaries Westlaw's editors write above court opinions to sum up each point of law. To teach its search tool which passages answer a legal question, ROSS had another company, LegalEase, write about 25,000 training memos, each pairing a legal question with passages from court opinions graded by how well they answered it. The memo writers used Westlaw headnotes to write the questions because, the court found, they offered "an easy way" to do it. The appeal covered 2,243 headnotes the trial court found were copied. The judges held that the headnotes are protected by copyright because Westlaw's editors chose which points of law to include and how to word them. The court opinions they summarize stay free for anyone to use.

ROSS argued its copying was fair use, the part of U.S. copyright law that allows some uses of copyrighted work without permission, such as for criticism, news reporting, teaching or research. The court disagreed. It said ROSS used the headnotes for the same purpose as Westlaw, to help people find relevant court opinions, and advertised its tool at prices "in line with" Westlaw's to win the same customers. ROSS could have worked from the court opinions themselves, which were freely available. "Unlike necessity, ease is not a justification for copying," Judge Montgomery-Reeves wrote for the three-judge panel. The court also found that a market for licensing headnotes as AI training material is growing fast, and that ROSS's copying took that chance away from Thomson Reuters.

The ruling is narrow. ROSS's tool did not write anything new; it only pointed users to passages from existing court opinions. The judges noted that the U.S. Justice Department, in a filing in the copyright case against OpenAI, argued that training AI that writes original text can count as a new kind of use. They said that reasoning does not apply here, because ROSS's tool could not write anything new and was built to replace Westlaw. Courts decide fair use case by case, so a ruling against one company's copying does not decide how other cases will come out, including the larger fights over chatbots.

The Association of American Publishers, the book publishers' trade group, which filed a brief backing Thomson Reuters, welcomed the decision as a "milestone" and said the question of AI training is being argued in about 150 U.S. cases. ROSS itself shut down its service in January 2021, saying the costly lawsuit had been heavy to bear. The appeal answered two questions the trial judge sent up before the case was finished, so it is not yet clear what ROSS may owe, or whether it will ask a larger panel of the court or the Supreme Court to review the decision.

More on AI policy and law