Suno has answered the second copyright suit from Universal and Sony with a statement that, read closely, gives the labels something they wanted: an admission that its v6 models were trained in part on songs made by Suno's own users. The company's wording lists licensed partner content, then "interactions including creations and preference signals" from its community, then its own accumulated know-how. Creations sit beside preference signals in that sentence, not inside them.
That matters because of what Suno said at launch two weeks earlier. Its product chief told Music Business Worldwide that v6 was built from the ground up on licensed material and user data, and the company clarified that the user data meant preference choices, not audio people had uploaded. Uploads were never really the question. The tracks users generate with the older models were, and this is the first time Suno has placed them in v6's training account.
The labels' complaint, filed in Boston on 18 September over 60,202 recordings, is built for exactly this. It argues that anything the earlier models produced carries the expression of the unlicensed catalogue they were trained on, so a new model taught from those outputs, or from users' preferences among them, inherits the infringement rather than escaping it. A second theory says v6 was distilled from its predecessors as a student learns from a teacher, with the same result.
Suno calls the claims flawed on the facts and the law, points to its deals with Warner, BMG and Believe, and says it exists so more people can make music. Its own launch post never described the training data at all. What the court now has is Suno's account of that data in its own words, and a dispute over whether a model can ever start fresh when its teachers did not.
Sources · 1