Carolina Gomez – We have officially entered the Age of AI. As artificial intelligence continues to transform industries, legal battles over its development are intensifying. Copyright owners across the nation are suing AI companies for illegally using their copyrighted materials to develop AI models. Among the plaintiffs are authors, visual artists, media companies, and music-industry giants. In response, many AI companies are asserting a fair use defense.
One of the first cases to test this defense was decided on Tuesday, February 11, in the U.S. District Court of Delaware by Judge Stephanos Bibas (“Judge Bibas”). The case, initially filed in 2020, involves Thomson Reuters (“Reuters”), a media company that owns the famous legal research platform Westlaw, and Ross Intelligence Inc. (“Ross”), an AI legal research startup. Reuters claimed that Ross illegally used its Westlaw materials to train its AI research platform, infringing on its copyrights. Ross, in turn, argued that its use was permissible under the fair use doctrine.
Westlaw’s database consists of complied case law, statutes, regulations, and proprietary editorial content, such as headnotes and a classification system known as the Key Number System. Reuters’ primary claim was that Ross infringed on Westlaw’s copyrighted headnotes and Key Number System by using them without permission to train its AI model.
Ross initially sought to license Westlaw’s content, but Westlaw declined due to competitive concerns, prompting Ross to find an alternative. Ross subsequently partnered with LegalEase Solutions, a third-party legal research company, to create training materials for its AI model. These materials, called Bulk Memos, were drafted by lawyers who compiled thousands of legal questions and answers using Westlaw’s headnotes and numbering system as references. Although Ross’ final product did not provide end users with access to Westlaw’s copyrighted material, its AI model was built on them, raising copyright concerns.
In 2023, both parties moved for summary judgment on various claims. Judge Bibas denied the motions for both sides and set a trial date for August 2024. However, before the trial date, he revisited the case, postponed the trial, and invited the parties to renew their summary judgment motions. Earlier this week on February 11, Judge Bibas ruled on those motions, granting partial summary judgment in favor of Reuters on its direct copyright infringement claim while denying Ross’ fair use defense.
A court evaluates whether something is fair use by analyzing the following four factors: (1) the purpose and character of the use; (2) the nature of the copyrighted work; (3) the amount and substantiality of the portion used; and (4) the market effects of the use. While courts analyze these fair use factors on a case-by-case basis, some factors have emerged as determinative or carrying more weight than others over the years. In this case, Bibas concluded that the first and fourth factors favored Reuters, while the second and third favored Ross, but “ultimately carried less weight.”
Under the first factor, courts assess whether the use is transformative, meaning it adds new expression or meaning to the original work. Ross argued that its use of Westlaw’s material was transformative because it served as an intermediate step in developing its software. In previous cases, intermediate copying was deemed fair use when it was necessary for achieving software compatibility. However, relying on the Supreme Court’s 2023 ruling in Andy Warhol Foundation for the Visual Arts Inc. v. Goldsmith, Bibas held that “Ross’ copying wasn’t ‘reasonably necessary to achieve the user’s new purpose.’” Unlike the software cases where copying was necessary for compatibility, Ross could have trained its AI model using other sources.
For the second factor, courts consider whether the copyrighted work is creative or factual and whether it is published or unpublished. Creative works receive stronger copyright protection, whereas factual works receive thinner protection and are more likely to qualify as fair use. Bibas found that while Westlaw’s headnotes and Key Number System are partially creative, they are primarily factual, making this factor weigh in Ross’ favor. He also noted that this factor rarely plays a significant role in fair use cases.
Under the third factor, courts analyze how much of the copyrighted work was used and whether the copied portions were central to the work. Even minimal copying can weigh against fair use if the copied portions make up the heart of the work. Bibas determined that because Ross used Westlaw’s materials only at the training stage and did not provide end users with direct access to Westlaw’s content, this factor favored Ross.
The fourth factor, which Bibas deemed the most important in his fair-use analysis, examines whether the infringement harms the current or potential market for the original work. Citing the Supreme Court’s ruling in Harper & Row v. Nation Enterprises, Bibas concluded that Ross’ product was “meant to compete with Westlaw by developing a market substitute.” By using Westlaw’s content to train its AI, Ross created a competing legal research tool, harming Reuters’ market position.
Bibas’ ruling has sparked debate among legal experts. Some argue that factors two and three should have carried more weight in his analysis and that his reasoning for concluding that the “tie” favored Reuters was underdeveloped. Another critique is that this opinion may have limited applicability to the dozens of pending AI cases, as this case involves non-generative AI, whereas most pending cases focus on generative AI.
Non-generative AI produces fixed outputs from specific inputs, while generative AI creates entirely new outputs based on the input information. Despite this distinction, the ruling will likely influence upcoming AI copyright cases. Not only is it one of the first decisions addressing AI and copyright, but it also sets a precedent for how courts may balance the fair use factors in the AI context.
Although a trial is still set to resolve additional claims involving over 5,000 of Westlaw’s headnotes, this ruling sends a clear warning to AI developers. If they can no longer rely on fair use as a defense, they may be forced to license copyrighted materials, driving up development costs and potentially limiting their ability to create new AI technologies.
Moreover, companies that have already built AI models using copyrighted materials without permission now face an increased risk of legal liability, as this decision could prompt more copyright holders to pursue litigation. At the same time, copyright holders may leverage this ruling to justify higher licensing fees, further restricting AI research and innovation—the exact chilling effects that copyright law seeks to prevent. As courts continue to grapple with AI copyright disputes, companies should proceed with caution, as fair use may no longer offer a reliable legal shield. Ultimately, Judge Bibas’ ruling has ushered in a new legal landscape in the Age of AI, where rapidly advancing technologies threaten to challenge and reshape traditional copyright principles.
