The first complaint was filed on Nov. 6 in the Northern District of California and runs on an inventory: 269 musical compositions that Eight Mile Style owns or controls, among them the publishing rights to Eminem’s catalog. The second, filed the same day in the same court, does not count songs. It describes a voice, and argues that Suno’s software can reproduce it on demand for anyone with a credit card and a sentence of description.
Both suits name the same defendant, and both arrive as AI music generation has settled into a paid subscription product rather than a demo. They are worth reading together because they test two different legal theories against the same technology, and only one of them depends on how the model was trained.
The Eight Mile Style complaint is the conventional case, and it is built to be expensive. It alleges direct copyright infringement, circumvention of technical protection measures under the Digital Millennium Copyright Act, and removal of copyright management information. It cites reporting from 404 Media that Suno used a service to pull millions of songs and lyrics from YouTube Music, Deezer and Genius, and it offers a demonstration: a prompt for “Lose Yourself” returned a track reproducing the lyrics of the Eminem song word for word. The publisher asks the court to impound or destroy unauthorized copies and to award damages that could exceed $40 million, a figure that tracks the statutory maximum for willful infringement across a catalog of that size.
The Presley complaint is the more revealing document. It is brought by rights holders connected to Elvis Presley’s estate, and it leans on a Tennessee statute passed in 2024, the Ensuring Likeness Voice and Image Security Act, which makes it unlawful to use software to replicate a performer’s voice without authorization. The law was drafted before the current generation of music models existed, and it was drafted to reach exactly the conduct the plaintiffs describe: a user types a description, and a model returns something that sounds like a specific dead singer.
That distinction moves the argument away from training data. Suno’s defense in the earlier litigation brought by the major record labels has rested on fair use, the claim that learning from recordings is transformative and that its output is new. Whatever a court eventually decides about that, a state statute aimed at voice replication does not require a plaintiff to prove anything about the training set. It requires a demonstration that the system produced a likeness, and that the likeness belonged to someone who never consented to it.
Suno’s public response has been consistent across both cases. The company said it exists to help people make original music rather than to trade on anyone’s name, that it prohibits prompts naming specific artists and songs, and that it screens uploaded audio and lyrics through third-party providers. Every clause of that statement describes a filter, and the plaintiffs’ theory is that the filters are an admission: a platform that blocks artist names has decided that artist names are the thing worth blocking.
The damages arithmetic differs in a way that favors the plaintiffs. Statutory damages hand a court a schedule to work from, and the schedule is generous when infringement is found to be willful. A right of publicity claim hands a jury an argument about commercial value instead, and the value of a voice that has been licensed for decades, in films, in advertisements, in a themed attraction, is not a number anyone has had to compute. The estate’s licensing business is the closest available benchmark, and it prices nostalgia by the year.
Behind both filings sits a question the industry has been deferring for two years: whether the licensing market for training data will be built by negotiation or by litigation. The major labels have signed deals with some AI music companies and sued others. Publishers have moved more slowly, partly because their catalogs are fragmented across thousands of titles and partly because the statutory damages may be worth more than the licenses. A complaint filed at the maximum is a bet that a court will price the catalog higher than a deal would have. For Suno, the near-term exposure is not a judgment but a calendar. It is the cost of defending several suits at once while raising capital and renewing licenses, plus the possibility that a court grants an injunction covering output rather than training. An injunction that requires a company to stop generating a category of sound is easier to enforce than one that requires it to unlearn something, and it is easier for a competitor to design around.
Neither filing will settle whether AI music is legal. They will settle what it costs, and to whom, which is a narrower question with a larger consequence for anyone who pays for a subscription. The industry has spent three years arguing about inputs. The plaintiffs this week are arguing about outputs, and outputs are the part a court can hear.
