What happened
The US Court of Appeals for the Third Circuit issued a precedential judgment upholding the district court's ruling in favor of Thomson Reuters against Ross Intelligence. The court rejected the fair use defense for copying copyrighted legal summaries to train a competing AI search tool. The decision, authored by Judge Tamika Montgomery-Reeves, is currently under seal but will bind district courts in the circuit once unsealed. The ruling affirms that using copyrighted works to create a direct competitor does not qualify as fair use, a position strongly supported by amicus briefs from the RIAA, NMPA, and major film studios.
The US Court of Appeals for the Third Circuit has upheld a lower court's decision in the long-running copyright dispute between Thomson Reuters and Ross Intelligence. The court ruled that Ross's copying of Westlaw's editorial summaries to train its AI-powered legal search engine did not constitute fair use. This judgment affirms the partial summary judgment granted by Judge Stephanos Bibas in February 2025, which had previously rejected Ross's fair use defense.
The appellate decision is a 32-page precedential opinion authored by Judge Tamika Montgomery-Reeves, joined by Judges Luis Felipe Restrepo and Emil Bove. Although the full opinion is currently under seal, the court has ordered Ross Intelligence to cover Thomson Reuters' costs on appeal. The parties have been given 10 days to propose redactions before the court decides whether to unseal the document. Once unsealed, this opinion will bind district courts within the Third Circuit.
The case originated in 2020 when Thomson Reuters sued Ross Intelligence for copying Westlaw's headnotes—short editorial summaries of legal points—to train a competing search tool. Ross shut down its platform in 2021 due to litigation costs. The appeal focused on two certified questions: whether the headnotes and West's Key Number System are original works, and whether Ross's use of them was fair use. The Third Circuit's ruling effectively answers the latter in the negative for this specific context.
The decision was bolstered by a joint amicus brief filed by the Recording Industry Association of America (RIAA) and the National Music Publishers’ Association (NMPA) on November 25, 2025. The brief argued that training AI models on copyrighted works to create competing services 'cannot qualify as fair use.' The RIAA and NMPA highlighted the scale of AI-generated content, citing Deezer's report of receiving over 50,000 AI-generated tracks daily in late 2025, a figure that rose to nearly 90,000 by June 2026.
Source details: musicbusinessworldwide.com ↗
Why it matters
This ruling provides the first federal appellate precedent specifically addressing the fair use defense in the context of AI training data for competitive products. By affirming that copying copyrighted material to build a rival service is not fair use, the decision significantly raises the legal risk for AI companies relying on unlicensed data. It validates the existence of a licensing market for training data, as evidenced by recent deals between AI firms and major music labels. This legal clarity may force AI developers to negotiate licenses or alter their data sourcing strategies, potentially impacting the cost and structure of future AI model development across various industries.
This ruling is significant because it is the first federal appellate decision to explicitly address the fair use defense in the context of AI training data used to create a direct competitor. While the specific case involved a non-generative search tool, the legal principle that copying copyrighted works to substitute for them is not fair use has broad implications for the AI industry. It challenges the common defense that AI training is a transformative use of data.
The decision validates the argument that a licensing market for AI training data exists and is growing. The RIAA and NMPA pointed to recent licensing deals between AI companies like Stability AI, Udio, and ElevenLabs with major music labels and publishers. By acknowledging these deals, the court undermines the argument that no market exists for licensing training data, thereby strengthening the position of copyright holders in future negotiations and litigation.
For AI developers, this precedent increases the legal and financial risks associated with using unlicensed copyrighted material for training. Companies may need to invest more in licensing agreements or develop alternative data sourcing strategies to avoid litigation. This could lead to higher costs for AI model development and potentially slower innovation in areas where high-quality, copyrighted data is essential.
The ruling also has implications for the music industry, where AI-generated content is increasingly flooding streaming platforms. By establishing a clear legal boundary against using copyrighted music to train competing AI models, the decision supports efforts to protect human artists from being 'unwittingly conscripted' into undermining their own markets. It may also influence the outcome of pending cases involving companies like Suno and Udio, which have made fair use a central part of their defense.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Impossibility results in algorithmic fairness (e.g. Kleinberg et al., Chouldechova) show what?
What to watch next
Monitor the unsealing of the 32-page opinion to understand the specific legal reasoning applied. Watch for the impact on pending cases involving companies like Suno and Udio, where fair use is a central defense. Observe whether other AI companies accelerate licensing negotiations with content creators in response to this precedent. Track the timeline for the summary judgment motions in the Suno and Udio cases, which are now scheduled for 2027.
The unsealing of the Third Circuit's opinion is a critical next step. The specific legal reasoning and any nuances in the ruling will provide more detailed guidance for lower courts and AI companies. The 10-day window for proposing redactions suggests the opinion may be released soon, offering a clearer picture of the court's stance on AI training data.
The impact on the Suno and Udio cases will be closely watched. Both companies are currently facing copyright lawsuits from major music labels, with fair use as a key defense. While the Thomson Reuters case involved non-, the principles established may influence how courts view the use of copyrighted music in training generative models. The summary judgment motions in these cases are scheduled for June 2027, but the Third Circuit's ruling could shape the arguments and strategies of the parties involved.
The growth of licensing deals between AI companies and content creators is another area to monitor. The RIAA and NMPA's emphasis on the emergence of a licensing market suggests that more AI companies may seek to negotiate licenses to avoid litigation. This could lead to new business models for AI training data and potentially higher costs for AI developers.
The broader implications for AI policy and regulation may also be significant. The ruling could influence legislative efforts to address AI copyright issues, providing a judicial precedent that supports the rights of copyright holders. It may also encourage other industries to pursue similar legal actions against AI companies that use their copyrighted works without permission.