Suno just released v6, a new music generation model trained exclusively on licensed music. The timing isn’t subtle. The company is currently defending itself against lawsuits from the Recording Industry Association of America, Universal Music Group, and Sony Music, all claiming earlier versions of Suno were trained on copyrighted songs without permission.

The new model represents a sharp pivot. Suno says v6 wasn’t trained on any music used for previous versions. It’s also promising not to generate songs in the style of specific artists without licenses, a feature competitors like Udio already offer.

But the lawsuits aren’t going away. They target Suno v3 and v4, which launched before the company struck licensing deals. The RIAA suit, filed in federal court in Massachusetts, alleges Suno scraped and reproduced thousands of copyrighted recordings to train those models. Universal and Sony filed similar complaints.

Suno has argued its training use qualifies as fair use under copyright law. That’s the same defense OpenAI, Anthropic, and other AI companies are making in their own copyright litigation. Courts haven’t ruled definitively on whether training AI models on copyrighted material without permission is legal, making these cases critical for the entire industry.

The licensing strategy

Suno’s shift to licensed training data puts it in a different position than most AI companies. While OpenAI and Google argue they can train on anything they can access, Suno is now paying rights holders directly. The company hasn’t disclosed terms or which labels it’s working with, but the move suggests it thinks the legal risk of unlicensed training outweighs the cost of deals.

That calculus makes sense for a music-specific product. Music rights holders have decades of experience litigating unauthorized use, and they’ve won consistently. The industry killed Napster, forced YouTube into licensing deals, and extracted billions from streaming services. Suno probably looked at that track record and decided it couldn’t win a war of attrition.

The question is whether licensing will work as a business model. Training AI models is expensive, and music licensing adds another major cost. Suno will need to convince users its new model is good enough to justify a subscription, even if competitors offer cheaper or free alternatives trained on unlicensed data.

What happens next

The lawsuits will proceed regardless of Suno’s new model. The complaints seek damages for past infringement, not just an injunction against future use. If the plaintiffs win, Suno could owe statutory damages of up to $150,000 per willfully infringed work. With thousands of songs allegedly used in training, that could mean hundreds of millions in liability.

Suno will likely argue it stopped the allegedly infringing behavior by launching v6, which could reduce damages even if it loses on liability. Courts sometimes show leniency to defendants who take corrective action. But the labels will push for maximum penalties to deter other companies from training first and licensing later.

The broader legal question matters more than Suno’s fate. If courts rule AI training requires licenses, every company building models will need to renegotiate access to data. That would shift significant leverage to copyright holders and could slow AI development in creative fields. If fair use wins, companies can keep training on whatever they can scrape, and licensing becomes optional.

Suno is trying to have it both ways: defending fair use in court while building a licensed alternative. It’s a hedge that acknowledges the law isn’t settled. The company gets to keep selling its product while the cases play out, and if it loses, it can point to v6 as evidence of good faith.

Whether that’s enough to avoid catastrophic damages won’t be clear for years. These cases move slowly, and appeals are likely no matter who wins at trial. In the meantime, Suno is betting users will pay for a model that doesn’t come with a potential shutdown notice attached.