Until now, AI companies seem completely impervious to copyright infringement lawsuits. Although many had attempted to challenge alleged copyright infringement committed by AI companies, none had been successful. It seemed as though AI occupied a legal gray area of copyright law – and there was nothing that anyone could do about it. However, a recent landmark case has shown that AI companies are vulnerable to copyright infringement lawsuits. This is the first victory of its kind, and many say that it will change the way AI companies operate forever. AI startups should keep this case – and its implications – in mind as they move forward. Company representatives may want to discuss these implications with an experienced technology lawyer.

Thomson Reuters Wins AI Copyright Infringement Case

Thomson Reuters is a Canadian multinational corporation that describes itself as a “world-leading content and technology company.” It provides news and “information-based tools” to journalists, legal professionals, and others. The company focuses heavily on information related to legal matters, regulatory issues, and taxes. It also offers software to accountants and CPAs.

One of Thomson Reuter’s flagship products is Westlaw, a digital legal research tool described by the company as the industry leader. Lawyers frequently use this platform for legal research, and it provides a database of case law, statutes, secondary sources, and other information that helps them build cases and plan strategies. The only real competitor in the modern era is LexisNexis, although studies indicate that Westlaw is far more popular.

In February 2025, the company made headlines after winning a landmark case in Delaware. Thomson Reuters was suing an AI startup called “ROSS,” alleging that the company accessed the Westlaw database and used it without permission to train its AI model. ROSS was in the process of creating a generative AI solution that would have been a direct competitor to the Westlaw database.

As most AI entrepreneurs know, training AI models can be challenging. To build these models, startups must give their AI models access to vast databases. This can be costly, but some companies have circumvented these costs by gaining access to certain databases without permission. Many AI startups have been accused of accessing pirated versions of books, prompting many authors to claim copyright infringement. However, these lawsuits have been largely unsuccessful.

Predictably, ROSS attempted to use the “fair use” defense. This is the same defense strategy used successfully by major AI companies, including those sued for accessing pirated books online. The strategy was successful because under current law, AI models do not count as “authors.” Instead, the responsibility for infringement falls upon anyone who might use the AI to generate works that violate existing copyrights. In other words, AI is simply a tool – and not a copyright violator in and of itself.
ROSS failed to use this strategy effectively. The court found that ROSS copied Westlaw headnotes to train its large language model in a way that did not constitute fair use. The judge in this case found that these headnotes were original – dismissing ROSS’s argument that this was merely “unprotected ideas and facts” with no market. The judge pointed out that the headnotes are not copied court documents because they involve “editorial judgment.”

In addition, the court found that ROSS intended to use these headnotes in a way that was commercial and non-transformative. Finally, the court concluded that this infringement caused “market harm” to Reuters, potentially affecting the Westlaw SaaS while undermining existing markets for AI-driven legal research. It is worth noting that ROSS is now defunct. This is the first major ruling for plaintiffs in AI copyright cases.

The Implications of This Decision

Some say that this decision could have major implications for AI startups in the future. Thanks to this ruling, startups may need to obtain licensing agreements before accessing proprietary datasets to train their large language models. These licensing agreements could be expensive or impossible to obtain due to the reservations of established companies. Westlaw and LexisNexis dominate their field and are unlikely to grant anyone access to their proprietary datasets. The same may be said for many other oligopolies in the tech world.

However, this all depends on how other courts interpret this decision. Some are less enthusiastic about the implications of this case, stating that it will not have much of an impact on other generative AI lawsuits that are still pending.

TechTarget points out that this case involved traditional and not generative AI, arguing that its implications are therefore limited. ROSS simply copied and pasted Westlaw headnotes to train its AI model. This straightforward copying was non-transformative, and this made it easier to rule in favor of Reuters.

If ROSS had transformed the headnotes with generative AI before training the large language model, the outcome might have been very different. Even if they had manually reworded these headnotes – even slightly – this could have been enough to avoid copyright infringement issues.

This provides some degree of hope for AI startups – including those already engaged in copyright infringement lawsuits. However, it is always important to understand how the fair use defense works – and a tech lawyer may be able to shed more light on this complex topic. Alternatively, startups today may want to take the “safe route” and simply train their models on licensed datasets.

Can a Technology Lawyer Help Me With AI Copyright Infringement Cases?

If your startup faces copyright infringement allegations, the same legal defenses you once relied on may no longer be reliable. This landmark case shows that AI companies can be successfully sued for copyright violations, and this is definitely something to keep in mind. With this precedent set, more AI companies may face the same legal threats in the near future. Aside from companies that already face these allegations, new startups may want to build their AI models on a foundation that is less vulnerable to legal action. These steps may be easier with guidance from a technology lawyer. Discuss this subject further during a consultation with John P. O’Brien.