Earlier last week, a pivotal US ruling regarding fair use in AI-related copyright litigation was delivered by Judge Stephanos Bibas in the case Thomson Reuters v. ROSS Intelligence, No. 1:20-cv-00613 (D. Del.). This decision represents a major milestone in AI-related legal battles, especially with respect to how copyrighted content is used to train AI models.
However, it’s important to note that while this ruling addresses the training of an AI model, it does not involve generative AI technology. As such, AI developers and deployers will need to closely monitor future rulings to determine whether the creation and distribution of AI-generated content will be considered fair use under copyright law.
Key Takeaways
- The court ruled that ROSS’s use of Thomson Reuters’ content to develop a competing AI-based legal platform does not qualify as fair use under the US Copyright Act.
- The court found that 2,243 Westlaw headnotes were directly copied, with substantial similarity.
- ROSS’s defenses—including innocent infringement, copyright misuse, merger, and scenes à faire—were rejected by the court.
- ROSS’s commercial use was heavily weighed against its fair use defense.
- The court reversed its previous decision to deny summary judgment on the issue of fair use.
Background
Thomson Reuters is the owner of Westlaw, one of the largest and most widely used legal research platforms in the US. Subscribers gain access to an extensive range of legal resources, such as case law, statutes, regulations, news, law review articles, and more. A critical component of Westlaw is its headnotes, which summarize the key points of legal opinions. Additionally, Westlaw includes the Key Number System, which organizes these legal opinions.
ROSS Intelligence, a competitor, sought to license Westlaw’s content for its own legal AI tool. When Thomson Reuters refused, ROSS acquired “Bulk Memos”—which were created using Westlaw’s headnotes—through a third-party legal services vendor. Upon discovering this, Thomson Reuters filed a lawsuit against ROSS for copyright infringement based on its use of Westlaw content to train its AI model.
Overview of the Case
The court granted partial summary judgment in favor of Thomson Reuters, addressing direct copyright infringement, fair use, and other defenses, while rejecting summary judgment motions from ROSS. The case was analyzed under the fair use framework established by the US Copyright Act (17 § USC 107), which includes four factors:
- The purpose and character of use (commercial vs. nonprofit educational purposes)
- The nature of the copyrighted work
- The amount and substantiality of the portion used in relation to the copyrighted work
- The likely effect of the use on the market for the original work
The court found in favor of Thomson Reuters on the first factor. In assessing ROSS’s use, the court determined that it was commercial in nature. Although ROSS argued that its use was transformative—since it allegedly “transformed” headnotes into numerical data for its AI system—the court disagreed. It noted that ROSS’s use did not offer a different purpose or character than Thomson Reuters’ original use. The court also rejected ROSS’s reliance on the “intermediate copying” doctrine, noting that previous cases involving computer code were not relevant, as they dealt with the necessity of copying to foster innovation. In contrast, ROSS’s use of the headnotes was not required to achieve its goal.
On the second factor, the court sided with ROSS, finding that although Westlaw’s content has some originality, it is not highly creative. The headnotes possess some editorial creativity, but the Key Number System is largely a factual compilation with limited creativity.
The court ruled in ROSS’s favor on the third factor, despite the number of headnotes used, since the material was not publicly available and did not include the headnotes from Westlaw. The court emphasized that the amount used should be considered based on what is made accessible to the public, not just what was copied. ROSS’s output did not contain the actual Westlaw headnotes, making the third factor inapplicable.
The most critical aspect of the case came with the fourth factor. The court ruled that ROSS could have developed its own product without infringing Thomson Reuters’ copyrights. In its assessment of the likely effect of ROSS’s use on Westlaw’s market, the court concluded that ROSS intended to directly compete with Westlaw and failed to prove otherwise. This factor, often considered the most crucial in fair use cases, heavily favored Thomson Reuters.
Looking Ahead
This ruling has significant implications for AI-related copyright litigation and the arguments surrounding fair use.
One of the key issues at stake is whether creating an AI model can be considered transformative and thus qualify for fair use—especially since AI models typically store their intelligence as numerical weights updated during training. Despite the advanced technology behind ROSS’s model, the court declined to find its use transformative, largely due to the competitive nature of ROSS’s goals.
The court’s decision also underscores that non-generative AI systems, like the one used by ROSS, may not be afforded fair use protection, even if their output involves uncopyrighted verbatim text. Although ROSS’s AI system produced uncopyrighted quotes from court opinions, the court still found no fair use.
This decision may set the stage for future cases involving generative AI technologies, such as large language models (LLMs). As the court emphasized, “factor four is undoubtedly the single most important element of fair use.” If the output of generative AI systems (e.g., AI-generated art or text) is substantially similar to the original work and harms the market for the original, fair use may be difficult to prove.
The key takeaway for AI developers and deployers is this: they must closely monitor the ongoing litigation surrounding AI and copyright, keeping an eye on how fair use will be applied to both training models and AI-generated output. Given the facts-driven nature of fair use, courts are likely to issue different rulings based on the specifics of each case, particularly where commercial use is involved.