Sony Music Entertainment has taken its fight with AI music generator Udio to a new level, filing a fresh lawsuit that alleges the platform infringed copyright across more than 30,000 Sony-owned songs. The complaint, filed in New York court on Monday, is notable not just for the sheer number of works cited, but for what it signals about how major labels are trying to frame generative music litigation: less as a vague “AI sounds like music” dispute, and more as a data-and-output infringement case grounded in specific recordings and specific claims of copying.
At the center of Sony’s filing is a list of tracks that spans decades and marquee artists—ranging from Elvis Presley’s Hound Dog to Beyoncé’s Say My Name and Harry Styles’ As It Was. Those titles are being used as concrete examples of what Sony says Udio generated in ways that violate copyright. Sony’s complaint also emphasizes that the cited set is only a fraction of what it believes Udio infringed, describing the list as “only a small portion” of the works at issue. In other words, the lawsuit is not presented as a one-off batch of problematic outputs; it’s positioned as evidence of a broader pattern.
This is the second major wave of label litigation targeting AI music tools. In 2024, Sony joined other major rights holders—including Universal Music Group and Warner Records—in suing Udio and Suno. That earlier case has already shaped the legal landscape by pushing the dispute into the realm of training data and discovery. Sony’s latest filing builds on that momentum, using the results of prior legal steps to sharpen its allegations and expand the scope of what it claims Udio did wrong.
What makes this latest move especially consequential is the way it reflects the labels’ evolving strategy. Early in the AI music debate, many arguments were framed around whether AI systems can be said to “learn” from copyrighted material without infringing, and whether the resulting outputs are sufficiently transformative to avoid liability. But as lawsuits progress, the focus tends to narrow: courts and plaintiffs increasingly want specifics—what was used, what was copied, and how the output relates to protected expression.
Sony’s complaint appears designed to deliver those specifics. By naming thousands of songs and tying them to alleged infringement, the filing tries to convert an abstract technological controversy into a measurable legal claim. The list itself functions like a map: it tells the court which works Sony believes are implicated, and it gives the plaintiff a foundation for arguing that Udio’s outputs are not merely “in the style of” in a general sense, but instead reproduce or closely track protected elements of particular recordings and compositions.
The Verge reports that Sony’s lawsuit references the broader dispute over how AI systems may be trained and how their outputs may relate to copyrighted recordings. That framing matters because it highlights the two pressure points that have dominated AI copyright cases: training and generation. Training is where plaintiffs argue the system absorbs copyrighted material without permission. Generation is where plaintiffs argue the system’s outputs are too close to protected works—either by reproducing recognizable elements or by effectively enabling users to obtain infringing copies.
In practical terms, Sony’s approach suggests it wants the court to treat Udio not as a black box that produces “new music,” but as a system whose behavior can be evaluated against known copyrighted works. The inclusion of widely recognized songs is likely intentional. When a complaint cites famous tracks, it reduces the burden of persuasion for the public and for the court: the alleged similarities are easier to understand, and the stakes feel immediate. But the legal work is still complex. Copyright law distinguishes between sound recordings and musical compositions, and it also distinguishes between ideas, styles, and protectable expression. A lawsuit that names thousands of songs is essentially asking the court to navigate those distinctions at scale.
There’s also a strategic reason Sony may be emphasizing volume. In many copyright disputes, the number of works cited can influence how a judge views intent, knowledge, and the plausibility of defenses. If a platform allegedly infringes a handful of songs, it can be argued as coincidence or edge-case behavior. If it allegedly infringes tens of thousands, the argument shifts toward systemic issues—suggesting that the platform’s training and generation pipeline is producing outputs that repeatedly intersect with protected material.
That systemic framing is reinforced by Sony’s statement that the list represents only a small portion of the works it claims Udio infringed. This language is doing more than setting expectations; it’s telling the court that Sony believes the problem is not limited to the examples it chose for the complaint. It’s also a signal to Udio that the litigation could expand further, depending on what discovery reveals and what additional evidence Sony can compile.
To understand why this matters beyond the parties involved, it helps to look at what generative music tools actually do. Users typically provide prompts—sometimes including genre, mood, instrumentation, tempo, and lyrics—and the system generates audio accordingly. The legal question becomes: when does that output cross the line from “learning patterns” into “copying protected expression”? Plaintiffs often argue that even if the system doesn’t reproduce a full track verbatim, it can still infringe by generating substantial similarity to protected elements, such as melody, harmony, lyrics, arrangement, or distinctive sonic features tied to a recording.
Defendants, by contrast, frequently argue that outputs are statistically derived and not direct copies, and that the system’s training process does not necessarily involve memorization of specific works. They may also argue that the output is sufficiently original, or that the user’s prompt drives the result in a way that makes the output independent of any particular copyrighted source.
Sony’s lawsuit is essentially betting that the court will accept that the alleged similarities are not accidental and that the platform’s behavior is legally actionable. By citing specific songs—rather than only describing general concerns—Sony is trying to make the case testable. The court can compare outputs to the cited works, evaluate the nature of the similarities, and decide whether the alleged conduct meets the legal threshold for infringement.
Another layer to this story is the timing. Sony’s filing comes after earlier litigation and after Sony gained access to Udio’s training data through discovery, according to reporting referenced in the provided article summary. Discovery is often where AI cases become real. Without discovery, plaintiffs can only speculate about what the model learned. With discovery, they can attempt to show that the training dataset included copyrighted material, that the system was exposed to specific works, or that the model’s internal representations correlate with protected content.
Even when discovery doesn’t produce a smoking gun, it can still reshape the case. It can help plaintiffs identify which categories of works were included, how the training process worked, and whether the platform’s documentation aligns with its actual behavior. For defendants, discovery can be equally damaging, because it can reveal gaps in compliance, weaknesses in filtering, or inconsistencies in how the company describes its training practices.
Sony’s new lawsuit suggests that the company believes it now has enough to justify expanding the list of alleged infringements. That expansion is also a reminder that these cases are not static. As plaintiffs learn more, they can amend complaints, add new claims, or file additional actions. The legal system allows for iterative litigation, and major rights holders appear prepared to use that flexibility.
There’s also a broader industry implication: if Sony’s allegations hold up, it could affect how AI music platforms operate even before any final judgment. Lawsuits can lead to settlements, changes in training practices, or restrictions on what models can generate. Even the threat of injunctions—court orders that prevent certain behavior—can force companies to adjust quickly. In the AI music space, where technology evolves rapidly, legal uncertainty can become a business constraint.
But there’s a tension here that makes the case particularly interesting. Generative music tools are often marketed as creative instruments. They allow users to explore ideas quickly, experiment with arrangements, and generate drafts that might later be refined by human musicians. If courts interpret copyright broadly in a way that treats many outputs as infringing, it could chill experimentation and push the industry toward licensing-heavy models. If courts interpret copyright narrowly, it could embolden platforms and encourage more aggressive training strategies.
Sony’s lawsuit is therefore not just about Udio. It’s about the boundaries of what “creative use” means when the underlying system is trained on copyrighted works and produces outputs that may resemble those works. The labels are effectively asking the court to draw a line that protects rights holders from unlicensed ingestion and from outputs that replicate protected expression.
At the same time, the case raises questions that go beyond the courtroom. How should similarity be measured in AI-generated music? What counts as “substantial similarity” when the output is probabilistic and varies with prompts? How do you evaluate infringement when the system can generate multiple versions of a song-like structure, some closer to a target work than others? These are not purely legal questions; they’re technical questions too, and they often require expert testimony.
Sony’s decision to cite a large number of songs may be partly aimed at addressing these measurement challenges. With many examples, Sony can argue that the pattern is consistent: the system repeatedly generates outputs that intersect with protected works. That consistency can help plaintiffs argue that the similarities are not random artifacts of a complex model, but rather the predictable result of training and generation mechanisms.
For Udio, the defense will likely focus on several themes: whether the outputs are actually copies of protected works, whether the system’s training process involved lawful acquisition or fair use-like reasoning, and whether the alleged similarities are sufficiently specific to constitute infringement. Udio may also argue that the complaint’s list is overbroad or that the cited songs are being used as proxies for general style resemblance. In many copyright cases, defendants try to narrow the dispute to specific, provable instances rather than broad claims.
The court will ultimately have to decide what level of specificity is required. Plaintiffs generally need to show ownership of the copyrights and copying of protected elements. Ownership is usually straightforward for major labels with catalog records. Copying is the harder part, especially when the defendant argues that the system does not store or reproduce exact