In a first-of-its-kind decision, a US District Court ruled against a fair use defense for an AI model. In Thomson Reuters Enterprise Centre GHBM and West Publishing Corp. v. Ross Intelligence, Inc., the United States District Court for the District of Delaware examined whether Ross Intelligence (“Ross”) improperly used copyrighted data to train its artificial intelligence (AI) model. This decision revised a prior 2023 summary judgment opinion and order from the same court.

Ross, a startup company developing an AI-powered legal research platform, utilized Thomas Reuters’ and West Publishing’s copyrighted material (e.g., headnotes, which summarize points of law and case holdings, and West’s editorial content and annotations) that were protected by copyright law as training data to train its AI legal research engine. The court granted summary judgment to Thomson Reuters, notably on Ross’s fair use defense for its use of copyrighted material.

Ross used Thomson Reuters’ headnotes as AI data to create a legal research tool to compete with Westlaw. Ross hired a third-party vendor, LegalEase, to develop training data in the form of “Bulk Memos” to train its AI model. LegalEase sold Ross 25,000 Bulk Memos, which Ross used to train its AI search tool. Thomson Reuters owns copyrights in West’s copyrighted material and alleged that Ross infringed upon its copyrights.

One of the issues to be decided was whether Ross actually copied. Actual copying means “the defendant did, in fact, use the copyrighted work in creating his own.” Actual copying can be proved directly or indirectly. Directly, by providing evidence that the defendant copied the work, or indirectly, by showing that the defendant had access to it and produced something similar.

The court found the language of the headnotes to closely track the language of the Bulk Memo question but not the language of the case opinion. Thus, the court found substantial similarity in the headnotes as it was so obvious that no reasonable jury could find otherwise.

Ross asserted several defenses to copyright infringement. First, Ross asserted innocent infringement, which merely limits the damages but does not limit liability. Second, Ross asserted copyright misuse. Ross claimed that Thomson Reuters misused its copyright. Third, Ross asserted the merger defense. Ross claimed that any ideas were so close to the expression that they merged with the expression, making it uncopyrightable. Fourth, Ross claimed the scène à faire defense. This defense posits that certain elements or scenes in a work are so fundamental, common, customary, obligatory, and unavoidable that they are not considered original or protectable under copyright law, so they can be freely copied or adapted. The court rejected all four of Ross’s defenses.

Ross then asserted the fair use doctrine as a defense to its copying. Four factors have to be met to use fair use as an affirmative defense: (1) the use’s purpose and character, including whether it is commercial or for nonprofit; (2) the copyrighted work’s nature; (3) how much of the work was used and how substantial a part it was relative to the copyrighted work’s whole; and (4) how Ross’s use affected the copyrighted work’s value or potential market.

The court found elements one and four in favor of Thomson Reuters, elements two and three in favor of Ross, and because elements one and four mattered more, the court found summary judgment for Thomson Reuters on fair use. The court denied Ross’s motions for summary judgment on copyright claims. The court further granted summary judgment to Thomson Reuters against Ross’s defenses of innocent infringement, copyright misuse, merger, and scène à faire. Partial summary judgment was granted for Thomson Reuters on direct copyright infringement for 2,243 specific headnotes.

This case is the first to rule against the fair use defense for an AI model. The court’s ruling strengthened copyright protection for editorial content and annotations. This case has set a precedent that companies cannot freely use copyrighted materials to train AI models. A proper license is required, especially for commercial use. This ruling emphasized that AI developers should exercise greater care in using copyrighted materials as training data. This case involved a non-generative AI, which may lead to different defenses being applied concerning generative AI.

There are many lawsuits against other companies currently pending in courts, and decisions in those cases may involve similar questions about the use of the fair use defense. The distinction between the training data and resulting outputs from generative and non-generative AI will likely be key to deciding any future copyright case involving AI.

About Sanjay Agrawal

Sanjay is senior counsel in Tressler’s Litigation Practice Group in our St. Louis Metro Area and Chicago, Illinois offices. He focuses his practice on various general defense litigation matters. Sanjay has effectively litigated cases in Missouri as well as in various federal and appellate courts. He is an experienced attorney with over 20 years of legal experience in litigation, appeals, licensing, commercial transactions, due diligence, client counseling, patentability, freedom to operate, cross-border transactions, M&A, and technology commercialization. Click here to read Sanjay’s full attorney biography.