In a ruling that may redefine the boundaries of intellectual creation in the age of machines, the Third Circuit Court of Appeals has determined that training artificial intelligence on copyrighted material does not constitute fair use. The case, pitting Thomson Reuters against legal AI startup ROSS Intelligence, asks an ancient question in a new form: who owns the raw material of thought, and what may others do with it? The court's answer — that commercial competition cannot shelter itself behind the doctrine of transformation — arrives at a moment when the entire AI industry has staked its fo
Third Circuit rules AI training on copyrighted material not protected fair use
Fair use may not protect AI training on copyrighted work
Mark
So the Third Circuit just said you can't use copyrighted material to train AI. Does that mean every AI company is breaking the law right now?
Mimi
Not necessarily. This ruling is specific to Thomson Reuters versus ROSS Intelligence. It doesn't automatically apply to every AI training scenario. But it does suggest that the fair use defense—which many AI companies have been relying on—may not be as strong as they thought.
Luke
Right, and we should be careful here. The court focused on commercial substitution: ROSS was directly competing with Thomson Reuters' own products. That's a specific fact pattern. We don't know yet how courts will treat AI training that doesn't create a direct market substitute.
Mark
But the RIAA and NMPA filed briefs supporting Thomson Reuters. That's a big signal, isn't it?
Mimi
It is. The music industry especially has been worried about AI companies using songs without permission. This ruling gives them a legal foundation to argue that their copyrighted works shouldn't be used in training without a license.
Luke
Though we should note: the opinion is sealed for now, according to some of the reporting. We're seeing the outcome, but not the full reasoning. That matters when we're trying to predict how other courts will apply this.
Mark
Sealed? Why would a federal appeals decision be sealed?
Mimi
Sometimes courts seal opinions temporarily for various reasons—pending clarification, ongoing proceedings, or other procedural matters. It usually gets unsealed eventually, but for now, we're working with the headline and the parties' statements.
Luke
Which means we don't have the court's detailed analysis of fair use doctrine. We know the result, but the reasoning that got us there is still behind closed doors. That's a real limitation on how much we can say about what this means for other cases.
Mark
So what happens next? Does every AI company need to start licensing content?
Mimi
That's the big question. This ruling applies to the Third Circuit, which covers Pennsylvania, New Jersey, and Delaware. Other circuits might rule differently. But if similar cases reach other appeals courts and they agree with the Third Circuit, then yes, the pressure to license content would grow significantly.
Luke
And we should say: we don't know yet whether ROSS or Thomson Reuters will appeal to the Supreme Court. That could change everything, or it could leave this as the law in the Third Circuit while other circuits go different directions.
The Pulse
- The Third Circuit has handed the AI industry a legal shock: the fair use shield it has long assumed would protect AI training on copyrighted text has been struck down in a binding appellate ruling.
- Copyright holders from music publishers to legal database giants are rallying together, sensing that this case is the lever that could give them control — and compensation — over how their work feeds the machines.
- At the heart of the tension is a substitution problem: ROSS built a product that competed directly with Thomson Reuters, and the court found that transformation cannot be claimed when the new creation simply replaces the old one in the marketplace.
- AI companies now face a landscape where licensing negotiations, synthetic data, or public domain constraints may replace the open-scraping model that has quietly powered the generative AI boom.
- The ruling leaves deliberate gaps — nonprofit use, partial excerpts, non-competing applications — suggesting that courts will spend years drawing finer lines around what this decision actually forbids.
In a ruling that may redefine the boundaries of intellectual creation in the age of machines, the Third Circuit Court of Appeals has determined that training artificial intelligence on copyrighted material does not constitute fair use. The case, pitting Thomson Reuters against legal AI startup ROSS Intelligence, asks an ancient question in a new form: who owns the raw material of thought, and what may others do with it? The court's answer — that commercial competition cannot shelter itself behind the doctrine of transformation — arrives at a moment when the entire AI industry has staked its foundations on a different assumption.
A federal appeals court has handed Thomson Reuters a decisive victory over ROSS Intelligence, ruling that training an AI system on copyrighted legal research does not qualify as fair use. The Third Circuit's decision rejects one of the AI industry's most relied-upon legal assumptions — that the transformative nature of machine learning shields companies from copyright liability.
ROSS Intelligence had argued that its AI, trained on Thomson Reuters' legal analysis, produced something genuinely new: a system capable of answering legal questions. Fair use doctrine has long protected copying that transforms original material into something serving a different purpose. But the court found that ROSS's commercial ambitions and its direct competition with Thomson Reuters' own products undermined that argument. When a new use substitutes for the original in the marketplace, courts have historically been reluctant to call it fair.
The case drew support from beyond the legal industry. The RIAA and the NMPA filed briefs backing Thomson Reuters, signaling that music publishers and record labels see this ruling as a potential bulwark against AI companies using songs and compositions without permission. Copyright holders across industries are watching closely.
The implications ripple outward. Generative AI systems depend on massive datasets — much of it scraped from the internet, including copyrighted books and articles — and the industry has operated largely on the assumption that fair use permits this. If that assumption collapses under legal scrutiny, companies may be forced to license content from thousands of rights holders, develop synthetic training data, or limit themselves to public domain material.
The ruling is not without ambiguity. The Third Circuit anchored its reasoning in the specific facts before it: full commercial competition, use of entire works, and clear market substitution. Courts in other circuits, or cases involving nonprofit research or partial excerpts, may reach different conclusions. For now, however, the decision marks a significant turning point — one that places the question of who owns the raw material of artificial intelligence squarely before the law.
A federal appeals court has sided decisively with Thomson Reuters in a copyright dispute that will reverberate through the artificial intelligence industry. The Third Circuit Court of Appeals rejected the argument that using copyrighted material to train AI systems falls under fair use—a legal doctrine that typically permits limited copying for purposes like criticism, commentary, or education. The ruling upholds a lower court victory for Thomson Reuters against ROSS Intelligence, a legal tech company that had trained its AI system on Thomson Reuters' copyrighted legal research and analysis.
The case turns on a fundamental question about how AI companies can legally source the vast amounts of text needed to build their systems. ROSS Intelligence had argued that its use of Thomson Reuters' copyrighted materials qualified as fair use because the training process transformed the original content into something new—an AI model capable of answering legal questions. Fair use doctrine has long permitted copying when the new work serves a different purpose or adds new meaning. But the Third Circuit rejected this reasoning, finding that the commercial nature of ROSS's use and the direct competition with Thomson Reuters' own products weighed heavily against a fair use finding.
The decision carries weight beyond the two companies involved. The RIAA, representing major record labels, and the NMPA, representing music publishers, both filed briefs supporting Thomson Reuters' position. Their involvement signals that copyright holders across entertainment and publishing see this case as a test of whether they can control how their work is used in AI training. The music industry in particular has grown increasingly vocal about AI companies using songs and compositions without permission or compensation.
What makes this ruling significant is its timing and breadth. Generative AI systems—the technology behind tools like ChatGPT and similar language models—require enormous datasets to function. Most of these systems have been trained on vast quantities of text scraped from the internet, including copyrighted books, articles, and other materials. Companies building these systems have generally relied on fair use arguments to justify this practice. The Third Circuit's decision suggests that argument may not hold up in court, at least not in the way the AI industry has been deploying it.
The court's reasoning appears to focus on the commercial substitution problem: ROSS's AI system directly competed with Thomson Reuters' own legal research products. When a new use serves as a market substitute for the original work, courts have traditionally been skeptical of fair use claims. The fact that ROSS was a commercial venture, not a nonprofit educational project, also worked against it. These factors combined to convince the appeals court that fair use did not apply.
The practical implications are substantial. If this ruling holds and spreads to other circuits, AI companies may need to negotiate licenses with copyright holders before training their systems, rather than relying on the assumption that fair use permits such use. This could slow AI development, increase costs, or force companies to seek permission from thousands of copyright holders—a logistically complex task. Alternatively, it could push the industry toward using only public domain materials or creating synthetic training data, though both approaches have limitations.
The decision also leaves open questions about how broadly it will be applied. The Third Circuit focused on the specific facts of the Thomson Reuters case: direct commercial competition, use of the entire copyrighted work, and a clear market substitution. Future courts might distinguish other AI training scenarios. A nonprofit research project using copyrighted material for training, or a system that uses only small excerpts, might receive different treatment. But for now, the ruling stands as a significant victory for copyright holders and a potential constraint on how AI companies can build their systems.
Notable Quotes
The court rejected the argument that using copyrighted material to train AI systems falls under fair use— Third Circuit Court of Appeals ruling
Want the full story?
Read the original at Google News ↗