Suno replaces its AI models with a new one trained on licensed music as copyright suits pile up

Suno unveiled Suno v6, a model family it says was developed using licensed data from labels and distributors including Warner Music Group, BMG, and Believe—and not the data used to train earlier versions. Paying users get base v6 and an experimental "wild" variant; everyone gets a faster mini model, with older models slated for retirement.

The startup settled with Warner last year and struck a BMG deal last month, but still faces suits from Sony, Universal Music Group, artists such as Jason Isbell, and users alleging security failures. The announcement landed a day after Suno admitted training on YouTube videos.

Product hooks include prompt-based section edits, text/image/video references, instrument separation, planned opt-in remixes with labels, plus recent watermarking and tiered download limits aimed at streaming fraud and mass low-intent uploads.

This is the licensed-corpus pivot the music-AI category has been telegraphing: keep shipping features, swap the training story, and hope settlements plus steerable tools redefine the company as a partner rather than a scrapers-with-melody problem.

Signal Desk's read: retiring old models is the tell. If prior weights were the litigation magnet, leaving them live would undercut the clean-room narrative. v6 is both a product upgrade and an evidentiary posture—"this system wasn't trained on that pile." Warner and BMG deals give Suno cover to talk revenue share and opt-in remixes; Sony and Universal suits mean the courtroom chapter is not closed just because the marketing chapter changed.

Jack Brody's line about derivative work creating more revenue for rights holders is the business thesis. It may be true for partners inside the tent. It does not automatically moot claims about how earlier models were built, nor the YouTube-training admission's timing. Watermarks and download caps are concessions to distributors worried about spam uploads; they also acknowledge that uncontrolled generation is an ecosystem hazard, not only a creator convenience.

The overstated claim would be that licensed training ends the copyright war. The understated one is product lock-in: once users lean on section editing, multimodal references, and label-blessed remix rails, switching costs rise even if legal cloud remains. With more than $819 million raised per PitchBook, Suno can afford the reset. Rivals without label paper will feel the squeeze first.

AI music generators have faced a wave of label litigation over allegedly unlicensed training data. Suno's path—settle and partner with some rights holders while litigating with others—mirrors a broader industry split between licensing desks and hardline infringement theories.