The United States District Court for Delaware ruled in favor of Thomson Reuters concerning its copyright infringement case against the now-defunct artificial intelligence startup ROSS Intelligence. This marks the first win in an AI copyright case. The ruling sets a precedent that using copyrighted materials to train AI models, especially for commercial and competitive purposes, may not qualify as fair use.
AI and Copyright: Thomson Reuters Secures Victory in AI Copyright Dispute with Now-Defunct AI Firm ROSS Intelligence
Background
Thomson Reuters is a Canadian multinational technology conglomerate with business interests in information services. It owns and operates Westlaw, a leading legal research platform that provides access to a vast database of legal resources, including proprietary features like headnotes and the West Key Number System. On the other hand, ROSS Intelligence was a legal tech startup founded in 2014 that developed an AI-driven legal research tool designed to enhance the efficiency of legal research by utilizing artificial intelligence to interpret and answer legal queries.
In May 2020, Thomson Reuters filed a lawsuit against ROSS Intelligence, alleging unlawful use of content from Westlaw to develop its own legal research platform. It specifically asserted that ROSS collaborated with LegalEase Solutions to access and copy proprietary content of Westlaw without authorization. These included its headnotes and the West Key Number System. This, according to Thomson Reuters, allowed ROSS to expedite the development of its competing product without investing the necessary resources to create original content.
Position of Thomson Reuters
• A direct infringement of copyright occurred since headnotes and Key Number System were copied. The use was non-transformative and commercial because ROSS was aiming to create a competing legal research product. It reproduced and created derivative works based on the editorial content of Westlaw.
• ROSS was earlier denied a direct license to Westlaw. ROSS later induced LegalEase to breach its contract by copying proprietary content using its Westlaw license. The copied materials were specifically used to train the AI model of ROSS. This resulted in a product that competed with Westlaw.
Defense of ROSS Intelligence
• The use of content from Westlaw was under fair use because it was for training an AI model. The resulting output was transformative. The product of ROSS did not directly compete with Westlaw. It was an AI tool designed to provide judicial opinions rather than generate new content based on public legal texts.
• Both headnotes and the Key Number System of Westlaw lacked the originality required for copyright protection. The texts in Westlaw are also public legal documents. No proprietary data was public. The lawsuit was an attempt to stifle competition and maintain a monopoly over legal research materials.
Court Decisions
Delaware District Judge Stephanos Bibas earlier dismissed the antitrust counterclaim of ROSS. The company alleged that Thomson Reuters violated federal antitrust law by unlawfully tying its search tool to its public law database. However, in September 2024, the court concluded that ROSS failed to provide sufficient evidence to support its claims.
Judge Bibas further granted summary judgment in favor of Thomson Reuters on the copyright infringement claims in February 2025. The court ruled that the editorial content of Westlaw, including headnotes and the Key Number System, is protected by copyright and that the use of this content did not qualify as fair use.
ROSS Intelligence was specifically found to have copied 2243 Westlaw headnotes. These were sufficiently original and creative to qualify for copyright protection. The use of the content was considered commercial and non-transformative. The court further underscored that the actions of ROSS harmed the market for the content of Westlaw.
Implications
The ruling sets a precedent that using copyrighted materials to train AI models, especially for commercial and competitive purposes, may not qualify as fair use. It underscores the importance of the commercial and transformative nature of use in fair use analyses. Non-transformative uses, especially those that compete with the original work, are less likely to be protected. Concerns were raised for AI companies that rely on copyrighted data for training their AI models. This may have more profound implications for models intended for non-generative applications.
Furthermore, the ruling underscores the importance of respecting intellectual property rights, even when developing innovative technologies Moreover, with regard to startups or companies seeking to implement AI in their businesses, the ruling highlights the challenges they may face when entering markets dominated by established players. This could potentially stifle innovation in certain sectors. Note that following the lawsuit, ROSS Intelligence ceased operations in January 2021, citing the financial strain caused by the litigation.
FURTHER READINGS AND REFERENCES
- Thomson Reuters Enterprise Centre GmbH et al v. ROSS Intelligence Inc., No. 1:2020cv00613 – Document 669 (D. Del. 2024). Available online
- Thomson Reuters Enterprise Centre GmbH et al v. ROSS Intelligence Inc., No. 1:2020cv00613-SB – Document 770 (D. Del. 2025). Available online
Photo Credit: Jason Zhang/ The Thomson Reuters building in Scarborough, on Kennedy Road/Adapted/CC BY-SA 3.0