Sony and UMG sue Suno again over v6 model copyright claims

By Splitfeed EditorialLast updated Sep 26, 2026, 19:00 UTC

Sony Music Entertainment and Universal Music Group filed a new lawsuit against AI music startup Suno in late September 2026, this time targeting the company's v6 model. The core legal argument is that v6 was trained on outputs generated by earlier Suno models, and that those earlier models were themselves built on unlicensed music sourced from YouTube and other platforms. The labels contend that this chain of derivation means the infringement from the original training data carries forward into the newer model, regardless of whether Suno used fresh copyrighted recordings directly.

The suit matters because it tests a legal theory that goes beyond the first generation of AI copyright cases. Most prior litigation focused on whether training an AI on copyrighted works constitutes infringement at the point of ingestion. This case adds a layer: whether training on synthetic outputs, if those outputs are themselves tainted by unlicensed source material, perpetuates liability through successive model versions. Sony and UMG are notable in this context because they declined the licensing agreements that other major labels signed with Suno, meaning they have kept themselves in an adversarial legal posture while the rest of the industry moved toward commercial accommodation. How courts treat the derivative-output theory could have significant consequences for how AI companies iterate on their models going forward.

None of the three camps had published reactions at the time this story was filed, so their positions can only be anticipated. Pro-AI voices would typically argue that training on model outputs is meaningfully different from copying original recordings, and that penalizing iterative development would chill innovation across the industry. Anti-AI voices would be expected to say this case confirms that laundering unlicensed training data through intermediate model generations does not cleanse the underlying infringement, and that the labels are right to press the point legally. A middle-ground perspective would likely acknowledge both the legitimacy of the labels' concern about accountability and the need for clearer statutory or judicial guidance rather than resolution through piecemeal litigation.

The argument that will decide the practical stakes is whether a court accepts the chain-of-infringement theory applied to synthetic training data. Any ruling on a motion to dismiss or on summary judgment in this new case, as well as any decision in the earlier Suno litigation still working through the courts, would give the first concrete signal about how far copyright liability can follow AI-generated material across model generations.