AI Pushes Copyright Limits—Court Pushes Back

AI Pushes Copyright Limits—Court Pushes Back

Artificial Intelligence lawsuits have been grabbing headlines for a few years now. One of the most common legal challenges is whether AI should be free to train on data that is protected by copyright and owned by third parties without first obtaining permission. Is this fair use?

Briefly, in US Copyright Law, fair use is a legal doctrine that allows limited use of copyrighted material without permission for purposes such as criticism, commentary, news reporting, teaching, scholarship, or research. Luckily, we now have an answer from the U.S. District Court in Delaware: using copyrighted works to train AI models without permission is not fair use.

Thomson Reuters owns the legal research database Westlaw, which includes headnotes (editorial summaries written by legal experts). These headnotes help attorneys and judges quickly understand case rulings. ROSS Intelligence, an AI-driven legal research startup, aimed to create a competitor to Westlaw.

In 2020, Thomson Reuters sued ROSS, alleging that ROSS used headnotes without permission to train its AI. ROSS initially tried to license Westlaw’s data but was denied. Instead, it used “Bulk Memos”—documents prepared by a third party that were derived from Westlaw’s headnotes. ROSS’s product, when prompted with a legal question, would produce a judicial opinion—but not a Westlaw headnote. ROSS was using the headnotes to train the AI model but was not reproducing the headnotes to the end user. ROSS argued this was fair use, but Thomson Reuters disagreed, leading to a legal battle.

In 2023, the court had initially determined that fair use was a matter for the jury to determine and denied early dismissal of the case before trial. Then, after further arguments and briefing, and after personally reviewing thousands of headnotes, the judge reversed this portion of the earlier ruling and instead held that this was decidedly not a case of fair use. The judge reasoned that because the headnotes were being used to train AI and create a tool to compete directly with Thomson Reuters there was no innovative or transformative use of the copyrighted headnotes. Moreover, Thompson Reuters may have been able to license its headnotes to other companies looking to train AI models or even use the headnotes to develop its own AI tools. Thus, no fair use.

The judge further noted that ROSS’s product was not a generative-AI model, meaning that the ROSS product was not creating new content, but outputting existing judicial opinions. However, this case will likely prove to be a bellwether for other cases involving generative AI.

The line between fair use and infringement will likely be tested further as AI evolves and these lawsuits make their way through the court system. For now, this case is a warning to AI developers: borrowing without a license can be costly.