The Recording Industry Association of America has filed lawsuits against generative music startups Udio and Suno, putting one of the most sensitive questions in AI directly in front of the music business: what happens when a model appears to depend on copyrighted work it was never licensed to use?
The dispute is not only about two companies. It is about whether generative music AI can be built on large catalogs of recorded music without permission, then offered as a product that can create songs on demand. If the cases move forward, they could become a major test for training data, fair-use arguments, and the future of AI music tools.
What the RIAA is alleging
The lawsuits filed by the RIAA target Suno and Udio, with Udio identified in the source as Uncharted Labs doing business as Udio. The core claim is that the companies used copyrighted recorded music at broad scale to train their music-generating AI systems.
According to the source, the lawsuits argue that these models were trained on music owned by labels and copyright holders without permission. The point is not simply that the companies built AI tools. The issue is the alleged source of the material used to make those tools work.
The source describes the lawsuits against Udio and Suno as extremely similar. Both center on the same basic concern: a generative music model needs a large amount of high-quality music to learn patterns, and the RIAA says these companies used protected recordings to supply that training material.
Why training data is the center of the fight
Generative music AI is presented to users as a system that can create new songs from prompts. But the source argues that these systems operate by matching prompts to patterns learned from training data and completing those patterns. That makes the contents of the training dataset a decisive issue.
If a model can produce music that closely resembles existing songs, the RIAA can argue that the original recordings were part of the training process. The source says the RIAA claims to have examples where songs owned by its members are being reproduced in altered but recognizable form by the models.
The article points to the difference between text models and music models. In the source's view, a text model may be able to argue that short excerpts came from reviews, previews, or other fragments available online. A music model faces a harder problem if it can reproduce substantial parts of a recognizable song, including the structure and musical content.
The examples named in the source include Jackson 5, Maroon 5, "Great Balls of Fire," and "Call Me Maybe." The legal relevance is straightforward: if a generated track is plainly based on a protected recording, it becomes harder to argue that the model never had access to the original work.
The fair-use argument and its limits
Udio and Suno, according to the source, told RIAA lawyers that they believe the material they ingested falls under fair-use doctrine. That matters because fair use is the concept companies often invoke when they use copyrighted works without direct authorization.
The source also notes that the companies and their investors have spoken openly about the copyright challenges involved in building a strong music generation model. Antonio Rodriguez of Matrix Partners told Rolling Stone: "Honestly, if we had deals with labels when this company got started, I probably wouldn’t have invested in it. I think that they needed to make this product without the constraints."
That statement is important because it frames licensing as a constraint the product was built without. In a lawsuit over unauthorized use, that kind of public discussion can become part of the broader factual picture around what the companies knew and how they made decisions.
The companies are also described as arguing that their systems are not intended to replicate copyrighted works. The source characterizes this as an effort to shift responsibility toward users, similar to how platforms may argue they are not liable for every copyrighted song a user adds to uploaded content. But the source suggests that argument may be difficult if the companies themselves used copyrighted music in training.
What could happen if the cases advance
The most immediate consequence could be forced disclosure. The source predicts that the companies may have to reveal training data and methods because those details are directly relevant to the lawsuits. For AI companies, that kind of discovery can be as significant as the final ruling.
If evidence shows that copyrighted material was misused, the source suggests several possible outcomes. The companies could try to settle, avoid trial, or face a fast judgment. The RIAA could also seek an injunction if it can persuade the court that Udio and Suno are causing major harm to copyright holders and artists.
An injunction would matter because it could stop the operation at an early stage of the trial. The source frames this as more plausible in music than in some text disputes because the allegedly generated works may be closer to complete recreations of known songs.
There is also a business question. The source argues that at least one of the companies might try to continue using legal or legal-adjacent music sources, but that doing so could reduce model quality by the companies' own standards for training data. If users leave because the product becomes weaker, the legal issue becomes a market issue too.
Why investors will be watching
The lawsuits also create a warning for investors in generative media. The source argues that if a hot generative media startup can lose a huge amount of value because of how its product was built, future investors may demand deeper diligence before funding similar companies.
That would change the incentives around AI music and other generative tools. Instead of treating training data as a technical detail, investors may need to examine whether the model depends on material that could trigger copyright claims.
The broader lesson is not limited to Udio and Suno. The RIAA lawsuits may show what happens when a generative AI company builds first and answers copyright questions later. For the music industry, the case is about protected recordings. For the AI industry, it is about whether the foundation of a model can become its biggest legal risk.