The U.S. Court of Appeals for the Third Circuit has ruled that using copyrighted material to train an AI system is not protected by fair use, handing Thomson Reuters a landmark victory in its long-running case against legal research startup Ross Intelligence. The decision is among the clearest appellate rebukes yet to the argument that large-scale AI training can be justified as transformative use under U.S. copyright law.
The ruling matters well beyond the parties involved. It arrives as courts across the United States are being asked to decide whether developers of generative AI models may ingest books, news articles, music, and other protected works without permission or compensation. For investors, the opinion adds a new layer of legal risk to a sector that has already been pressured by copyright claims, licensing negotiations, and uncertainty over the durability of AI business models built on vast data extraction.
Fair Use Narrowed
The Third Circuit's reasoning is significant because it cuts against one of the core defenses used by AI companies: that training is merely a technical process, not a market substitute for the original work. In this case, the court found that Ross's use of Thomson Reuters' legal content was not sufficiently transformative to qualify as fair use. That conclusion could influence how judges evaluate similar disputes involving model training, retrieval systems, and other AI products that depend on copyrighted source material.
The opinion also underscores a broader judicial skepticism toward claims that the scale of AI innovation alone can override copyright protections. While the technology sector has argued that model training is essential to machine learning and therefore should receive broad legal latitude, the court's decision suggests that necessity is not the same as immunity. The ruling may encourage rights holders to press harder for licensing fees, injunctions, or damages in cases involving text, audio, images, and code.
Market Implications Rise
For global markets and equities, the decision is likely to be read as both a legal and commercial inflection point. AI developers, cloud providers, and platform companies have spent heavily to build and deploy systems that rely on enormous corpora of data. If courts increasingly require permission for training inputs, the economics of AI could shift toward paid content access, tighter data governance, and higher compliance costs.
That prospect is especially relevant for publicly traded firms with exposure to AI infrastructure, content distribution, and enterprise software. Companies that have marketed themselves on rapid AI deployment may now face questions about the provenance of their training data, the scope of their indemnities, and the possibility of future liabilities. Media owners, publishers, and database operators, by contrast, may gain bargaining power as the legal environment becomes more favorable to copyright enforcement.
The ruling also arrives at a moment when investors are already reassessing the gap between AI enthusiasm and monetization. A more restrictive copyright regime could slow product development in some areas, but it may also accelerate the emergence of licensed data markets and premium partnerships. In that sense, the decision does not merely constrain AI firms; it may help define the next phase of the industry's commercial architecture.
Copyright Pressure Builds
The case has drawn attention from across the creative and content industries, which have increasingly challenged the use of protected works in AI training. The Third Circuit's decision may embolden similar claims from publishers, music companies, and other rights holders seeking to establish that training on copyrighted material requires authorization. It could also shape settlement dynamics in pending litigation, as defendants confront a less favorable appellate precedent.
Still, the ruling is not the final word on AI and copyright. Other courts may interpret fair use differently, and the Supreme Court could eventually be asked to resolve the conflict if appellate decisions diverge. But for now, the Third Circuit has handed copyright owners a powerful argument: that copying works to train an AI system is not automatically transformative, and that innovation does not erase the need for permission.
For the AI sector, the message is stark. The legal foundation for training on copyrighted material is weaker than many developers had hoped, and the cost of building compliant systems may rise accordingly. For markets, that means the next phase of AI growth may be shaped as much by licensing law as by computing power.