Suno's New Model Only Sings With Licensed Music. That's an Admission the Old One Was a Liability, and Your Back Catalog Didn't Get the Memo.
Suno announced its v6 model family this week, and the headline detail is simple: it's trained only on licensed music from Warner Music Group, BMG, and Believe, artists who opted in, catalogs Suno actually has rights to. The company says v6 doesn't use the same training data as its earlier models. Read that carefully and it's not really a product announcement, it's an admission. The old models were built on something Suno itself no longer wants to stand behind, and every track a creator generated on those earlier models is still sitting out there, in videos, in ad libraries, in someone's small business, built on a legal foundation the company just quietly moved away from.
What actually shipped
Suno settled with Warner Music Group last November and struck a deal with BMG last month; v6 is the first model built specifically on the licensed catalogs those settlements unlocked. Sony Music and Universal Music Group are not settled, they're still litigating. Separately, Munich's Regional Court ruled in July for GEMA, the German performing rights society, against Suno, prohibiting the company's use of six specific compositions across both training and output, ordering disclosure of how extensively that material was used, and holding Suno liable for damages. That ruling didn't touch Suno's whole catalog, but it's a concrete, court-tested precedent that training on unlicensed music carries real financial and legal consequences, not theoretical ones.
Put those together and the sequence is clear: label sues, label settles, licensed model gets built from the settlement, model gets marketed as the safe new default. It's a rational path for Suno to take. It's also a pattern that anyone building on a generative platform should recognize, because it isn't unique to music.
The part that doesn't get fixed retroactively
Here's what v6's launch doesn't do: it doesn't relicense anything a creator already generated on the old models. If you built a product, a YouTube channel, a background-music library for video editors, an app that generates jingles for small businesses, on top of Suno's earlier output, that catalog's legal footing hasn't changed because Suno shipped a cleaner model going forward. The tracks you already have are still tied to whatever training data produced them, and Sony and Universal's active lawsuits, plus the GEMA ruling, are all about that older material, not v6's.
This is the uncomfortable gap in "we fixed it going forward" announcements: they read as good news, and for new generations they are, but they say nothing about exposure you already accumulated. If your business depends on a library of AI-generated music built before this week, the honest move is figuring out which model generated which track and whether any of that material touches catalogs currently in litigation, not assuming v6's announcement retroactively cleans up your position.
Why this isn't really a music story
Suno is the case study this week because the timeline is public and unusually well-documented, settlements, an active court ruling, a named replacement model, all lined up close together. But the underlying pattern applies to every generative platform trained by scraping first and licensing later: image generators, video generators, voice cloning tools, code-completion models trained on public repositories with murky license terms. The generative AI era's default go-to-market has been "ship first, negotiate the licensing exposure after it's a business," and Suno's v6 launch is what it looks like when that negotiation starts actually landing, label by label, court by court.
If you've built a product around any generative platform's output, the question worth asking isn't "is this specific platform in trouble." It's "what's this platform's training data story, and what happens to my product the day that story changes." Some platforms (fully licensed from day one, or trained entirely on data they own) don't have this exposure. A lot of the popular ones do, and most solo operators building on them have never checked which category their vendor falls into.
The honest take
I think Suno is doing the right thing here, settling with labels, building a genuinely licensed model, and being public about the training-data distinction between old and new versions is more transparency than most generative AI companies have offered under similar legal pressure. This isn't a company getting caught and lying about it, it's a company getting caught and visibly correcting course. That's worth crediting even while pointing out the gap it leaves for people who built on the earlier versions.
Where I could be wrong: it's possible the existing catalog risk is smaller in practice than it looks on paper, plenty of AI-generated content from earlier, more legally ambiguous models has circulated for years without individual creators facing consequences, and the legal exposure may end up landing almost entirely on Suno itself rather than flowing downstream to people who used the tool. I can't rule that out. But "it probably won't come after you personally" isn't the same as "the underlying legal question got resolved," and the second one is what actually matters for anyone deciding whether to keep building on a platform's older output.
What I'd actually do
If your business includes AI-generated music, images, or video from any platform, spend an afternoon documenting which model or version generated which asset in your catalog, and which of those platforms have active litigation over their training data. That's not paranoia, it's the same due diligence you'd do before licensing stock footage from an unfamiliar source. And when you evaluate a new generative tool going forward, ask the licensing question before the quality question. A model that sounds slightly worse but is actually licensed cleanly is a better foundation for a business than one that sounds better today and might not exist in its current legal form in a year.
Author
Lukas
@lukcombinator