Suno announces audio watermarking, fingerprinting and download limits for tracks generated on the platform. Updated guidelines explicitly prohibit using a real person’s voice or image without permission, deceptive audio passed off as authentic, and bulk exports to streaming platforms.

The announcement comes as Suno fights on three legal fronts. The RIAA, coordinating with Universal and Sony, has sued over training on protected material. A German court ruled in favor of GEMA, establishing that Suno trained its models on copyright-protected tracks and that the American fair use doctrine doesn’t apply. A class action in Massachusetts follows a data breach in November 2025 that affected 55 million users, according to Have I Been Pwned.

For those building with AI, the mechanism matters more than the music. Suno didn’t choose watermarking on principle: it’s engineering it now because a German court ruled that American fair use doesn’t cover its training. The pattern is recurring in generative AI: lawsuits first, then controls become infrastructure. For those shipping models to production, compliance needs design from the start. It arrives from outside as a constraint, and those who don’t anticipate it find themselves chasing rules written by those who preceded them in court.

In detail

Suno is one of the leading AI music generators: give it a text description and it returns a complete track with vocals and arrangement. It raised $400 million in a Series D round in June 2026, and now finds itself at the center of one of the most important legal battles in generative AI.

What was there before. Until now Suno operated with generic guidelines and no structured system to track its outputs. Anyone could generate tracks and upload them to streaming platforms. In March, a man was convicted for uploading hundreds of thousands of AI-generated tracks and fraudulently collecting $8 million in royalties. The case showed that the platform had no tools to identify its own outputs once they left its perimeter.

What changes. Three new measures. Audio watermarking embeds an inaudible signal in generated tracks so other platforms can identify them as created by Suno. It’s unclear yet whether Suno will use an existing system like Google’s Synth ID or a proprietary one; the company hasn’t responded to TechCrunch on this. Fingerprinting, through an agreement with Musixmatch and its Sentinel system, detects lyrics and audio that reproduce protected works. Finally, the new download policy limits bulk exports to streaming platforms, targeting those generating thousands of tracks daily.

The German court ruling. The turning point is the German court’s decision favoring GEMA, Germany’s licensing agency. The court established two things: that Suno trained its models on copyright-protected tracks, and that American fair use doctrine doesn’t apply to a company operating this way. The second part is most relevant for the industry. Fair use permits using protected material for transformative purposes (creating something new and not directly competitive). But a Suno investor admitted that generated tracks compete directly with those of human artists, weakening the transformative use argument.

Where sources diverge. The Decoder and TechCrunch agree on the main facts but differ on details. The Decoder reports that Suno claims it excluded artist names from training metadata and never allowed prompts targeting specific artists. TechCrunch doesn’t mention these details but flags the November 2025 data breach (55 million users according to Have I Been Pwned), which The Decoder doesn’t cite. Together the two sources paint a more complete picture than either alone.

Limits of what we know. The announced measures are intentions for now. Suno hasn’t specified when watermarking goes live or what concrete download limits will take effect. The policy on artist names in training metadata applies to future models, not those already trained. And the data breach, with its class action, is a separate front concerning data security rather than copyright.

The implication for builders. Suno is a textbook case of governance emerging from lawsuits, not principles. For those developing products with generative models, the lesson is operational: watermarking, output traceability and limits on bulk use should be designed before launch, not after the first ruling.