Sony and Universal Say This AI Music App Copied Their Songs — Again
Sony Music and Universal Music Group have sued AI music company Suno again, this time over its new v6 system. They say v6 violates copyright. The case focuses only on v6, not older models, and is their second case against Suno.
The suit was filed in federal court in Massachusetts. The filing is 45 pages long, according to details reported on Sept. 18 Variety. Coverage describes it as a second lawsuit after the labels' original 2024 case The Hollywood Reporter.
The fight is over how v6 was taught. Sony and UMG say v6 learned from songs users made with older Suno models. Those older models, they say, had learned from unlicensed music taken from YouTube and other sources The Verge. Suno has no license from the labels.
The labels call this "model laundering." Think of a photocopy of a photocopy: if the first copy was made without permission, copying that copy does not fix it. Sony also says Suno used distillation, where a new AI system is taught to copy the behavior of older systems.
Suno disagrees. Jack Brody of Suno said v6 was built from scratch with a new set of data including user data The Verge. The company has not licensed music from Sony and UMG, but says v6 uses different training data than the older models in the first case.
On June 24, 2024, Sony, Universal and Warner Records sued Suno and Udio for copyright infringement Reuters. Their claim was that training AI on copyrighted recordings without permission breaks the law. The new case extends that claim to a newer system trained partly on AI-made songs from the old system.
Udio took a different path. Universal and Udio settled and agreed to build a new licensed AI music platform together Universal Music Group. They plan to launch it in 2026. Suno has no such deal with Sony or UMG.
The broader context here is what carries over between versions. If version one learned from unlicensed music, does version two inherit the problem when it learns from version one. The labels say yes. Suno says starting over with new data resets the chain.
In my view, this matters beyond music. Many AI tools now learn from past outputs and user activity. If courts say those outputs carry the legal problems of the first training data, companies must track data, filtering and permission more closely each time. That would also give music owners power over future models, not just the first one.
Worth flagging for buyers is the split now forming. One company settled with Universal to build a licensed platform. The other faces a second lawsuit over whether new training fixes old problems. Licensed music costs more and covers less. Other systems may do more, with more legal risk.
In practical terms, the case turns on details not yet public. What music was in the older training data. How much v6 data came from user songs. How the copying method worked. Whether filters limited memorization. Clearer rules could help musicians and startups build new music tools with permission.
