For At Least One AI Developer, Fair-Use Defense Fails Against Copyright Infringement Claims

In Thomson v. Ross, the court concluded that the developer’s unauthorized use of the third-party copyrighted works to train an AI tool did not qualify as fair use. That said, unlike many other cases pending against AI developers, the AI tool in this case is not generative AI (Gen AI) and therefore its influence on other pending AI cases remains uncertain.

Background

The plaintiffs, Thomson Reuters Enterprise Centre GmbH and West Publishing Company are the creators and copyright owners of Westlaw, a well-known legal research platform. Westlaw comprises notable features, including its West Headnotes, which summarize case holdings and other important legal issues raised in cases. ROSS Intelligence, Inc., an AI developer that created a legal research search engine, used headnotes in the data set for its AI tool. Specifically, Ross purchased “Bulk Memos” from LegalEase that were built from the headnotes, which Ross then used in its training sets. Subsequently, on May 6, 2020, the plaintiffs filed a complaint against Ross for copyright infringement and tortious interference with contract.

As the case progressed through litigation, each party moved for summary judgment on the copyright infringement claim and fair-use defense. A judge in the District of Delaware issued an opinion on February 11 on the parties’ motions for summary judgment.[1] Notably, the court granted the motions for summary judgement on the issue of fair use in the plaintiffs’ favor.[2]

Court Rejects Fair-Use Defense

Fair-use is an affirmative defense against copyright infringement that allows a party to use copyrighted works in certain circumstances, without the permission of the copyright owner. In deciding that the fair-use defense is not available to Ross, the court considered the following fair-use factors:

Factor 1: Purpose and Character of Use

This factor ultimately asks whether a work is commercial or transformative. The court found this factor in favor of the plaintiffs, and ultimately found that Ross’ use was commercial, which disfavors fair use. The court noted that Ross was using the headnotes in its training set for a tool that would compete with the plaintiffs’ product, which is evidence of the commercial nature of Ross’ product.

Also important to the analysis was the fact that Ross’ tool is not Gen AI because it provides users with judicial opinions that have already been written. Further, the headnotes are not part of the output of Ross’ search engine — Ross’ use occurred at an intermediate step, but such copying was not reasonably necessary for Ross to create their legal research tool. Looking at the broader purpose and character of Ross’ use, the court found that Ross’ use was not transformative because it used headnotes to more easily create a legal research engine that would compete with Westlaw.

Factor 2: Nature of the Copyrighted Work

In analyzing the nature of the copyrighted work, courts assess the degree of creativity of the work, with increased protection given to more creative works. Here, the court found this factor in favor of Ross. Though headnotes have enough creative elements to be subject to copyright protection, they are relatively less creative than many other works. The court considered this factor to be less important than others.

Factor 3: Amount or Substantiality of the Portion Used

Courts also look to how much of a copyrighted work was used by the alleged infringer, and how substantial this part was relative to the work as a whole. For factor three, the court noted that the output of Ross’ product, which is what is available to its consumers, does not include the headnotes. This is significant because it is the amount and substantiality of the copyrighted work that the infringer makes publicly available that matters for this factor, not simply the percentage of copyrighted work the infringer uses.

Factor 4: Effect on Value or Potential Market

After acknowledging that market harm is the most important factor when weighing the fair-use factors, the court weighed the likely effect of Ross’ use of the headnotes on Westlaw’s current market and potential derivative markets. Under this factor, the court considers any potential and likely public benefits of the copying. Here, the court noted that the relevant market is legal-research platforms, and the potential derivative market is data to train legal AI tools. The potential for market harm is high because Ross’ product is meant to be a substitution to Westlaw in the marketplace, and this is not outweighed by public benefit because judicial opinions are already freely available, and there is no public right to the headnotes.

Key Takeaway

Although the court found only two of the four fair-use factors to be in the plaintiffs’ favor, the court did not weigh each factor equally. Factors one and four weighed most heavily on the analysis, and the plaintiffs ultimately prevailed after the court found both in their favor. The case will continue to proceed through litigation, but Ross will not be able to rely on fair use to defend against the plaintiffs’ copyright infringement claims.

For other plaintiffs alleging copyright infringement against AI developers, this case demonstrates that plaintiffs may be able to successfully defend against developers raising a fair-use defense. It is still possible, however, that AI developers in other cases may be able to successfully defend against copyright infringement allegations by raising fair use as an affirmative defense. In this case, the outcome largely depended on the fact that the court gave significant weight to the first and fourth factors, but little weight to the second factor, delineating how courts may weigh the fair-use factors but not necessarily how all courts will analyze the factors, particularly as facts change.

For example, this court indicated that it decided this issue only for non-Gen AI, which was relevant in determining that Ross’ use of the copyrighted work was not transformative. Due to Gen AI’s ability to generate new and original content, a court may be more likely to find the AI developer’s use to be transformative, which would weigh in the developer’s favor in finding fair use.

The jurisprudence surrounding copyright, fair use, and AI will continue to evolve as more courts perform these balancing tests. The AI, Metaverse, & Blockchain industry group at ArentFox Schiff will continue to monitor these legal developments.

[1] The court granted partial summary judgment in favor of the plaintiffs on the copyright infringement claim with a respect to a number of headnotes, but the issue with respect to the remaining headnotes will go to trial. The question of which headnotes remain covered by existing copyrights will also be decided at trial.

[2] The court had issued a prior opinion on these issues in 2023, largely denying the plaintiffs’ motion for summary judgment. In the February 11 opinion, Judge Bibas revises the 2023 ruling.