One of the many ongoing cases involving copyright laws and AI generative modules and programs involves plaintiffs, Thomson Reuters Enterprise Centre, GMBH, and West Publishing Corp., against a now-defunct business — Ross Intelligence, Inc. Ross Intelligence is accused of copyright infringement when it attempted to create an AI-driven alternative to one of Reuters’ online products, Westlaw Publishing. Westlaw is an online-based research database mostly used by legal firms. The core of the database contains cases, statutes, rules, and regulations for the federal and State governments, various regulatory agencies, and, increasingly, cases and statutes from around the world. These government-created and owned materials are not copyrightable and Reuters does not claim copyrights on those materials.

However, a key part of the Westlaw search program involves what are called “Headnotes.” These are created by Reuters employees and attempt to summarize various points in the cases. They are also linked by topics and numbers so that, in theory, one can find cases involving the same topic and topic number, even though that case might be from a different State or decided 50 years ago. These Headnotes are, without question, created by human original authorship and are copyrightable.

The gist of Reuters’ complaint was that Ross Intelligence scraped from the internet and accessed a great portion of the Westlaw database — including the Headnotes — to train Ross Intelligence’s AI program and module in an effort to create an AI-based competitor to Westlaw. There are nuances in every case, but these are the basic claims being made in most of the pending and ongoing AI copyright cases.

Ross Intelligence defended, with the main thrust being the doctrine of “fair use.” Under U.S. Copyright law, there is a major exception to copyright infringement, which is called “fair use.” This is the basic defense that many AI firms are using in these lawsuits to defend against claims of copyright infringement.

In what might be a “big” legal victory, on February 11, 2025, the judge in the case, Judge Stephanos Bibas, rejected all “fair use” arguments made by Ross Intelligence. See media reports here and here. For Judge Bibas, the key was the fact that Ross Intelligence was attempting to compete with Reuters/Westlaw by developing a market alternative. Courts generally look at four factors in determining “fair use,” and one factor is whether the alleged infringer has a commercial or profit motive. In this case, Judge Bibas found that Ross Intelligence did have that motive, and that motive predominated over the other factors.

Whether the legal victory is “big” depends on many things, such as whether other judges around the country follow Judge Bibas’ reasoning and whether other defendants will be deemed to have a high-profit motive like Ross Intelligence. Of course, in the end, the U.S. Supreme Court (or Congress) will have to step in to resolve these complicated issues. In the meantime, plaintiffs in these cases have a win to celebrate

The Four Fair Use Factors and Why They Mattered in Thomson Reuters v. Ross

Judge Bibas’s analysis in Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc., No. 20-613-JLH (D. Del. Feb. 11, 2025) is instructive not just for its outcome but for its careful application of the four-factor fair use test. The ruling provides a roadmap for how courts may analyze similar AI copyright claims going forward.

Factor One: Purpose and Character of the Use

The first fair use factor asks whether the use is transformative — that is, whether the new work adds new meaning, expression, or message to the original — and whether the use is commercial. AI companies have consistently argued that training is transformative because the purpose of the copying (extracting statistical patterns) is fundamentally different from the purpose of the original works (communicating information to human readers). Ross Intelligence made a similar argument.

Judge Bibas was skeptical. While he acknowledged that some transformative purpose existed, he found that the primary purpose of Ross Intelligence’s use was to build a commercial legal research product that would directly compete with Westlaw. The commercial and competitive nature of the use weighed heavily against fair use on the first factor. The Supreme Court’s guidance in Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508 (2023), that the commercial nature of a use is particularly significant when the use and the original serve the same or overlapping markets, reinforced this conclusion.

Factor Four: Market Effect

The fourth factor — the effect of the use on the potential market for the original — is often considered the most important, and it was central to Judge Bibas’s analysis. Ross Intelligence was attempting to build a product that would substitute for Westlaw, capturing market share and licensing revenue that would otherwise flow to Reuters. This is precisely the type of market harm that copyright law is designed to prevent. Where a use supplants the market for the original work rather than creating a new market, courts consistently deny fair use.

The contrast with cases like Authors Guild v. Google, Inc., 804 F.3d 202 (2d Cir. 2015) — where Google’s book-scanning project was found to be fair use in part because the digital index served a different market than the original books — illustrates why competitive purpose matters so much under the fourth factor.

Implications for Broader AI Copyright Litigation

The Thomson Reuters ruling is the first major merits decision in the AI copyright wave, and it will influence, though not bind, other federal courts. Several implications deserve attention:

• Competitive purpose is a critical variable. AI companies whose products directly substitute for the plaintiff’s original content — as Ross Intelligence’s legal research AI competed with Westlaw — face heightened fair use risk under the Thomson Reuters framework. General-purpose language models may be distinguishable on this point, but not necessarily in all contexts.
• The case concerned copyrightable elements only. Judge Bibas carefully limited his analysis to the Headnotes, which are original human-authored works, and expressly declined to hold that the underlying cases and statutes — which are government works in the public domain — could be the basis for copyright claims. AI companies frequently argue that their training data consists largely of non-copyrightable facts and public domain materials; the specific composition of the training dataset will be dispositive in each case.
• The ruling does not resolve output infringement. Whether AI-generated outputs can themselves infringe copyrighted works that were part of the training data remains an open and actively litigated question. This is addressed in other pending cases, including the New York Times litigation.

Next Steps in the Litigation

Because Ross Intelligence dissolved during the litigation, the case proceeded against the former company’s estate. The ruling resolves the fair use defense on summary judgment, meaning the case may now proceed to trial on damages — potentially including statutory damages under 17 U.S.C. § 504(c), which can range from $750 to $30,000 per work infringed, and up to $150,000 per work for willful infringement. Given the volume of Headnotes allegedly copied, potential damages exposure could be substantial.

Copyright holders whose works may have been used to train commercial AI products should assess their legal options now. Contact the Copyright and AI attorneys at Revision Legal or visit our copyright practice page to discuss your rights.

Contact the Copyright and AI Attorneys at Revision Legal

For more information, contact the experienced Copyright and AI Lawyers at Revision Legal. You can contact us through the form on this page or call (855) 473-8474.