Why This Matters
Sony Music’s move to identify more than 30,000 songs in its copyright case against Udio marks a significant escalation in one of the entertainment industry’s most closely watched fights over generative artificial intelligence. What began as a broad accusation that AI music platforms trained on protected recordings is now becoming a more granular battle over specific tracks, specific datasets and the evidentiary trail behind how AI systems learn to make music.
The case centers on allegations that Udio used copyrighted sound recordings without authorization to train its music-generation technology. Sony, alongside Universal Music Group and Warner Records, sued Udio and fellow AI music company Suno in 2024, arguing that the platforms could not have achieved their ability to mimic musical styles, structures and vocal textures without ingesting large volumes of commercially released music.
Sony’s latest step is notable because it attempts to move the dispute from theory to documentation. According to the company, material produced during discovery was reviewed with audio fingerprinting tools, leading Sony to identify tens of thousands of additional tracks it believes were present in Udio’s training materials. The label is now seeking to expand the scope of its claims to include those works.
That matters because copyright cases involving AI often hinge on what the companies can prove was actually used. It is one thing for rights holders to argue that an AI model appears to have absorbed the language of popular music; it is another to present a list of recordings allegedly detected in the underlying material. If the court allows Sony to add the songs, the potential scale of the litigation grows dramatically.
The financial stakes are equally large. Copyright law can carry substantial statutory damages if infringement is proven, and the addition of thousands of works could increase exposure for an AI company already operating in a high-risk legal environment. Even if the matter never reaches a final judgment, the expanded list could influence settlement pressure, licensing negotiations and investor confidence across the AI music sector.
For artists, songwriters, labels and publishers, the dispute also touches a deeper concern: whether generative AI companies can build commercial tools on decades of recorded music without striking deals with the people and companies that made those recordings valuable. The labels argue that music catalogs are not raw material for unrestricted technological experimentation. AI companies, meanwhile, have generally contended that training systems on existing works may be lawful under principles such as fair use, particularly when the output is not a direct copy.
Industry Context
The lawsuit arrives at a pivotal moment for the music business. AI-generated songs have moved from novelty to commercial threat with surprising speed, giving casual users the ability to create polished tracks in seconds. Platforms such as Udio and Suno have drawn attention for producing songs that can evoke familiar genres, eras and production styles, raising alarm among rights holders who see a new category of competition built on their own catalogs.
The major labels have spent the past year drawing a firm line between licensed AI experimentation and what they describe as unauthorized mass ingestion of recordings. Their position is not anti-technology in the abstract; the same companies are exploring AI tools for marketing, restoration, localization and fan engagement. But they are seeking to establish that generative music platforms need permission before training on copyrighted recordings at scale.
This is part of a broader wave of AI copyright litigation across entertainment and media. Authors, visual artists, news organizations, actors and music companies have all challenged how AI developers acquire and use training data. The central legal questions remain unsettled: Is copying works into training datasets infringement? Can training be considered transformative? Does the output market substitute for the original works? And how much transparency should AI companies be required to provide?
Music presents an especially sensitive test case because sound recordings are highly identifiable. Audio fingerprinting, long used to detect songs on streaming platforms and user-upload services, gives labels a technical tool that may be more precise than the methods available in some other creative fields. If plaintiffs can demonstrate that particular recordings appeared in training materials, courts may be forced to confront AI training practices in unusually concrete terms.
The business backdrop is also important. The music industry has spent two decades rebuilding itself after piracy and the collapse of physical sales, with streaming transforming catalogs into long-term revenue engines. For rights holders, generative AI represents both a new opportunity and a possible repeat of earlier disruption: a technology that scales quickly before licensing norms and compensation structures are in place.
Udio, which gained rapid visibility after launching to public users, is part of a new generation of AI companies trying to turn creative generation into a mainstream consumer product. The appeal is obvious: users can create songs from prompts, experiment with styles and produce content for social platforms without hiring musicians or producers. But that same accessibility is precisely what worries labels, who fear a flood of AI-made tracks could dilute the market for human-made music and exploit recognizable creative signatures without payment.
What Happens Next?
The immediate question is whether the court will permit Sony to expand its claims to include the additional songs it says were identified through discovery. If the request is granted, Udio would face a broader and more detailed complaint, potentially requiring a more extensive defense around how its model was trained and what data was used.
Discovery will remain central. The labels are likely to press for more information about training sources, data pipelines and internal decision-making, while Udio may challenge the methodology used to identify the recordings and argue that the alleged presence of songs does not automatically establish liability. Expect both sides to rely heavily on technical experts who can explain how audio fingerprinting works, what it can prove and where its limits may be.
The case could also accelerate licensing conversations outside the courtroom. If AI music companies believe litigation risk is growing, they may pursue deals with labels and publishers to secure access to catalogs under controlled terms. Conversely, rights holders may use the pressure of pending lawsuits to shape the price and conditions of future AI licenses.
For now, the dispute is less about one platform than about the rules of the road for an emerging creative economy. The outcome could help determine whether AI music develops through negotiated partnerships with rights holders or through courtroom battles over what technology companies can use without permission. Either way, Sony’s expanded song list signals that the music industry is preparing to litigate the AI question track by track, not just in broad principle.