Introduction
The recent ruling in the case of Thomson Reuters v. Ross by the US District Court for the District of Delaware has significant implications for copyright and fair use, particularly concerning AI developers [2]. The decision highlights the challenges AI companies face when using copyrighted materials for training purposes and sets a precedent that may influence future litigation in this area.
Description
A recent ruling by the US District Court for the District of Delaware in the case of Thomson Reuters v. Ross has raised significant concerns regarding copyright and fair use [4], particularly in relation to AI developers. The court found that ROSS Intelligence [7], Inc.’s unauthorized use of over 2,243 copyrighted headnotes from Westlaw to train its legal research AI model constituted direct copyright infringement. Thomson Reuters [1] [2] [3] [4] [5] [6] [7] [8], which owns Westlaw [2], argued that ROSS’s actions violated copyright law after ROSS attempted to license Westlaw’s content but was refused due to competitive concerns [2]. In response, ROSS hired LegalEase Solutions to generate training data by creating bulk memos using Westlaw’s headnotes [2], which led to the lawsuit.
The court determined that ROSS’s use of the headnotes did not qualify as fair use, emphasizing that the use was commercial and directly competed with Thomson Reuters’ services, significantly impacting the market [5]. In its analysis of the four statutory fair use factors, the court concluded that the first factor weighed heavily against ROSS, as its use was not transformative; it served the same function as Westlaw’s system [8]. Although ROSS argued that its use was non-commercial and did not involve the headnotes in its final product, the court noted that ROSS could have achieved its objectives without copying the headnotes [2]. While factors two and three were found to slightly favor ROSS [2], the court ultimately ruled in favor of Thomson Reuters on the fourth factor [2], highlighting that ROSS’s product significantly affected the potential market for AI training [2], which encompasses both legal research platforms and the market for data to train legal AI models [2]. The court dismissed ROSS’s claim that the public benefit of its product outweighed the unauthorized use of Thomson Reuters materials [2].
This ruling is particularly significant as it comes from a Federal Judge, whose decisions can set important precedents in the US legal landscape [1]. It suggests that plaintiffs in similar copyright infringement cases against AI developers may have a stronger position in contesting fair-use defenses [7], although outcomes may vary based on specific case facts and the nature of the AI technology involved [7]. The fair use analysis is inherently fact-specific [6], influenced by the purposes [6], types of content [6], training methods [5] [6], and outputs involved [6]. Future cases may involve more protectable expressions [6], complicating fair use arguments [3] [6], particularly when the defendant’s use directly competes with the plaintiff’s business [6].
The implications of this ruling extend to ongoing AI copyright litigation, influencing other cases [6], such as those involving music publishers and Anthropic PBC [6]. The court dismissed several defenses raised by ROSS, including innocent infringement and copyright misuse [6], which may resonate with defenses used by other AI companies [6]. Given the complexities surrounding copyright infringement liability in AI [6], parties should carefully consider contract terms related to liability for potential infringement [6]. AI service customers should scrutinize indemnity clauses in their agreements [6], as the ruling may affect negotiations [6]. Companies developing proprietary software with AI should limit the use of third-party copyrighted content in training [6], as this could lead to infringement [6], particularly if it results in competing products [6].
While other AI developers may attempt to differentiate their cases from Thomson Reuters v [6]. Ross [2] [3] [4] [5] [6] [7] [8], the outcomes of generative AI copyright cases remain uncertain [6], indicating that fair use may not universally protect all AI defendants [6]. Courts will likely prioritize market realities over technological advancements in AI copyright disputes [3], closely examining whether AI training serves a competitive commercial purpose [3]. Numerous lawsuits are ongoing [6], with copyright owners pursuing various infringement theories against AI platforms for training on copyrighted works or through the outputs generated [6]. Developers are advised to emphasize novel applications that do not compete directly with training data sources and to explore early licensing strategies to mitigate infringement risks [3]. Monitoring evolving standards and developing licensing frameworks for AI training can strengthen market-harm arguments [3], as even internal use of protected material may lead to liability if linked to commercial competition [3]. Ultimately, transformation alone cannot supersede copyright’s primary function of protecting investments in creative content from infringement [3], particularly by competitors of the copyright holder [3].
Conclusion
The ruling in Thomson Reuters v. Ross underscores the complexities of copyright law in the context of AI development. It serves as a cautionary tale for AI companies, emphasizing the need for careful consideration of copyright issues when using third-party content for training purposes. The decision may influence future legal strategies and negotiations, encouraging developers to seek licensing agreements and explore non-competitive applications to mitigate infringement risks. As the legal landscape continues to evolve, AI developers must remain vigilant and proactive in addressing copyright challenges to protect their innovations and investments.
References
[1] https://www.brownejacobson.com/insights/copyright-and-ai-ruling-with-broad-implications
[2] https://www.lexology.com/library/detail.aspx?g=e10aedfd-d742-434d-a305-1978b35f4ce7
[3] https://www.jdsupra.com/legalnews/fair-use-doctrine-falters-in-6959162/
[4] https://project-disco.org/intellectual-property/scholars-agree-opinion-in-thomson-reuters-v-ross-should-be-disregarded/
[5] https://www.womblebonddickinson.com/us/insights/alerts/thomson-reuters-emerges-victorious-ai-copyright-battle
[6] https://www.jdsupra.com/legalnews/does-training-an-ai-model-using-6580351/
[7] https://www.jdsupra.com/legalnews/for-at-least-one-ai-developer-fair-use-2242550/
[8] https://www.livelaw.in/tech-law/us-court-rules-in-favour-thomson-reuters-westlaw-holds-that-ross-intelligence-infringed-copyright-for-ai-training-286025
