Sony and Universal file second lawsuit against Suno over alleged 'model laundering'

Why This Matters

• The lawsuit tests a novel legal theory: that training a new model on the outputs of an infringing model does not reset liability but simply "launders" the original infringement.
• A ruling against Suno could reshape how AI music companies train future systems and whether they can rely on cascading, self-generated training data.
• The case is part of a broader wave of litigation over whether using copyrighted works to train AI systems constitutes infringement.

Background

Record labels and AI music companies have been locked in a widening legal fight over the use of copyrighted recordings in training datasets. Suno, which lets users generate original-sounding songs from text descriptions, has faced claims that its models were built on unlicensed music. The company has acknowledged in earlier court proceedings that it scraped audio from YouTube. The new complaint targets v6 specifically, arguing that even if Suno changed its approach, the new model inherited the allegedly infringing character of its predecessors. The "model laundering" argument is a relatively untested legal theory that asks courts to follow the chain of training data from one model generation to the next.

Key Perspectives

Sony and Universal Music Group: The labels argue that Suno's v6 remains infringing because its training data descends from earlier models trained on unlicensed recordings. They characterize the practice as laundering infringement rather than curing it. Suno and the generative AI sector: AI developers have generally argued that training on existing material is transformative and lawful, and Suno has previously acknowledged scraping while defending its practices. Its formal response to this particular complaint was not detailed in the reporting available. Critics and skeptics: The "model laundering" theory is untested, and courts could struggle to determine how much protected expression actually carries over from one model generation to the next. It remains an open question whether tracing the genealogy of training data matters or whether only a model's final output is relevant.

What to Watch

• Whether Suno moves to dismiss the new complaint or seeks to consolidate it with the earlier litigation.
• How courts treat the "model laundering" theory — specifically, whether training on synthetic model outputs can be treated as a fresh act of infringement.
• Whether other rightsholders who have not licensed their catalogs to Suno follow Sony and UMG's lead with their own lawsuits.