AI companies accessing copyrighted materials to train AI systems have used the Fair Use Doctrine as a defense ever since, well, ever since AI systems accessed copyrighted materials to train AI. And ironically, the recent Delaware case, Thomson Reuters v. Ross Intelligence addressing whether the use of copyrighted material is in fact, subject to the fair use defense, involves legal research and a bunch of lawyers.
Thomson Reuters owns and operates Westlaw, which is one of the largest and most reputable legal research resources. One of the hallmark elements of Westlaw is the use of headnotes, which are summaries of key points in legal cases. Westlaw also uses a “Key Number System” that is proprietary to Westlaw. Ross Intelligence, a competitor, created a legal research platform powered by artificial intelligence. In the development of their platform, Ross initially requested a license from Thomson Reuters so that it could use Westlaw's content to train Ross' AI system. Reuters said no.
As a workaround, Ross was able to get the training data from a company called LegalEase. LegalEase provided Ross with “Bulk Memos”, which are collections of legal questions and answers prepared by real human lawyers. The lawyers were instructed not to copy headnotes word for word however, many of the lawyers used those exact headnotes as a reference which resulted in many of the of the Bulk Memos being derived from Westlaw headnotes. This training data allowed Ross to train its AI and develop a competing research platform.
In May 2020, Reuters brought a lawsuit against Ross, alleging
that Ross had infringed on Reuters' copyright by using its
content to train a rival legal research tool. In 2023 the District
Court in Delaware denied most parts of Reuters' motion for
summary judgment, however, the judge eventually concluded his
initial ruling did not address all relevant issues. Both Reuters
and Ross were allowed to renew their summary judgment
arguments.
Reuters requested summary judgment on copyright infringement and
dismissal of related defenses; Ross sought summary judgment
declaring non-infringement and a fair use ruling. As to the fair
use question, the court ruled in favor of Reuters:
The Fair Use Doctrine
The fair use doctrine is a legal principle that allows limited use of copyrighted material without needing permission from the copyright owner. It is intended to balance the rights of copyright holders with the public's interest in accessing and using creative works for purposes such as education, research, and criticism.
Under U.S. law, Fair Use is
determined based on four factors:
Purpose and Character of the Use: Is the use
commercial or non profit or educational? Is it transformative (i.e.
does it add new expression or meaning to the original work).
The Nature of the Copyrighted Work: Is the work factual or
creative, with factual works being more likely to be considered
fair use.
The Amount and Substantiality of the Work Used: This
examines the quantity and importance of the portion used in
relation to the entire work. Using a smaller portion is more likely
to be considered fair use.
The Effect on the Market: What is the impact of the use on
the market value of the original work. If the use reduces the
market for the original work, it is less likely to be considered
fair use.
The Court found that:
1. Ross' use of the Westlaw content was both commercial and non transformative. In other words Ross used the content to create a competing legal research tool. Ross argued that the headnotes were employed only at a certain stage of the product's development and not actually shown to end users. The court rejected that argument first because Ross copied written content (other cases addressing this issue relate to code), and second that the copying was not required for innovation or a distinct purpose—Ross had simply reproduced headnotes to develop a competing product.
2. While the original content more factual (it was legal research after all), the Court ruled in favor of Ross on this factor.
3. The Court looked at both the quality and quantity of the work that was used by Ross. The test is whether the “heart” of the work has been copied and the Court found in Ross' favor here as well. Since the final product did not actually display Westlaw's headnotes to users and headnotes were only used internally to train the AI tool.
4. The Court looked at how the alleged infringement would affect the copyright holder's market. This factor clearly favored Reuters as Ross' platform was a market substitute for Westlaw, and therefore, clearly undercut Westlaw's market share. The Court noted that Ross could have developed its own training materials—either independently or with LegalEase—without using copyrighted headnotes.
End Result: Each side won on two of the four factors, however Ross's fair use argument was ultimately rejected, and Reuters prevailed. In practical terms, the Court determined that in this case, fair use is not a valid defense when copyrighted content is used to train AI tools without authorization.
Why This Matters: This ruling is a significant win for copyright owners. It reinforces the control copyright holders have over their material and may lead to new revenue possibilities through licensing arrangements with AI developers. It also imposes stricter requirements on AI development and emphasizes the importance of vetting training datasets. To avoid legal issues, AI developers are best advised to either use public-domain materials or secure appropriate licenses. Note though that Judge Bibas states that there is a distinction between this kind of AI and a generative AI tool. While the rationale in this case could influence lawsuits involving generative AI, there is room for courts to be more open to a fair use defense when there is a stronger argument for “transformative use” (factor one).
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
