Musician Opposes Stability AI Dismissal Push in Training Lawsuit

Photo Credit: Stability AI

Want to opt out of AI music generators’ training datasets? Good luck: Stability AI is aggressively seeking to dismiss the infringement lawsuit filed by musician Anders Manga, who’s firing back against the push.

We broke down the “Welcome to Darkness” artist’s copyright complaint in detail closer to 2026’s beginning. Long story short – and there’s also a contractual element in play here – North Carolina-based Manga is accusing Stability AI of training Stable Audio on his work without authorization under a tie-up with AudioSparx.

The way the plaintiff tells the story, he sought to remove his projects from AudioSparx “months before the public launch of Stable Audio.” But the music library, which is likewise a defendant, allegedly “denied that request” and proceeded to forward the relevant efforts to Stability AI for training in any event.

Another removal demand (spearheaded this time with the assistance of counsel) then failed to bring about the desired result for Manga, who said that the allegedly unauthorized training was “reflected in AudioSparx’s own AI royalty statements.”

Evidently, all these occurrences set the stage for litigation – and the initially mentioned dismissal attempt, centering on the court’s alleged lack of jurisdiction over Stability AI.

According to the latter defendant, Manga “alleges no contacts between Stability AI and North Carolina” – where the suit was submitted – save his “own presence” therein. Meanwhile, Stability emphasized its incorporation in Delaware and “principal place of business in California.”

As for the plaintiff’s retort, Manga refuted the argument by citing Stability’s active registration status, registered agent, registered office, and subscription sales in the Tar Heel State.

Additionally, Stability pushed for jurisdictional-grounds dismissal by underscoring the case’s focus on AI training, which purportedly didn’t occur in North Carolina.

“Stability attempts to cabin this case to ‘training only’ occurring elsewhere,” Manga responded. “But Stable Audio is a single commercial ecosystem: Stability allegedly copied Plaintiff’s works to train the model, then monetizes that trained model by selling subscriptions that provide access to, and derive their value from, the trained system.

“The training is the input; the subscription is the commercial access. The two cannot be separated for purposes of the ‘related to’ inquiry,” he continued.

Closing with a brief look at AudioSparx’s distinct dismissal push – in brief, the entity pointed to the above-highlighted 2015 agreement with Manga – the plaintiff explored the absence of explicit AI training permissions in the contract.

“The contract’s catch-all must be read in light of the enumerated human-listening end uses, uses in which recordings are delivered, performed, or synchronized for human consumption, not copying for training data where the music is not heard by humans,” Manga penned.

“The Agreement also does not mention AI, datasets, or model training, and it contains no ‘now known or hereafter devised’ clause extending the license to future technologies,” he proceeded.

Regarding what comes next, the presiding judge today signed off on a one-week response-deadline extension for Stability, which, with the consent of Manga, now has until March 9th to submit its reply brief.