AINEWS
Two major labels are suing Suno, and a German rights body just won a court ruling against it, so the AI song platform is moving first on a standard the industry hasn't picked.
Suno, the AI song generator that turns text prompts into full tracks, is going to start tagging every song it produces with an audio watermark. The move lands less as a transparency gesture than as a play to set the de facto industry standard before a German court, two major record labels, or a regulator picks one for them.
The company, led by co-founder and CEO Mikey Shulman and valued after a $400 million Series D round in June, announced on Thursday that it will audio-watermark and fingerprint every track made on its platform. It also signed a deal with lyrics provider Musixmatch to use that company's Sentinal copyright-detection system, and updated its community rules to explicitly prohibit "deceptive audio presented as real" and "using a real person's voice or likeness without permission."
The new tools will pass disclosure to listeners. They will not pass judgment on whether a track should exist in the first place, which is the part rights-holders care about most. Shulman's blog post framed the watermarking layer as giving "artists and platforms" disclosure options, not as a way to adjudicate infringement.
Watermarking is one of the few technical interventions the music industry broadly agrees is desirable. The hard question is which standard the industry converges on: Google's SynthID, the audio-tagging system already used on YouTube uploads; an open alternative; or a proprietary system built in-house. When TechCrunch asked which technology Suno would adopt, the company declined to say.
That silence matters because the legal pressure on Suno is no longer hypothetical. The platform is a co-defendant with Universal Music Group and Sony Music Group in an RIAA-coordinated lawsuit over training data and output. Late last month, a German court ruled in favor of GEMA, the country's main rights-licensing body, finding that Suno broke copyright rules. And in November 2025, 404 Media reported a data breach that revealed internal training practices: Suno allegedly scraped YouTube, Deezer, and Genius to build its models, regardless of those platforms' terms of service.
Each of those fronts pushes toward a different answer. Labels in the RIAA suit want a standard their own systems can read. The German ruling treats copyright as a matter for a court, not a code path. And the 404 Media reporting pulls the harder underlying question into the open: whether the training data was lawfully obtained at all.
Suno is also planning a download policy that would bar mass distribution of its output on streaming platforms. The company did not detail how that would work, which platforms it would cover, or how it would be enforced. A blanket no-redistribution rule would require streaming platforms to identify and reject Suno output, which depends on the watermark being readable across the entire distribution chain. That loops back to the same question Suno is not yet answering: whose standard?
The Musixmatch deal is the part most likely to be read as a one-off partnership. It is also the most concrete. Musixmatch's Sentinal system is already deployed across the lyric provider's catalog and is designed to flag matching content against known rights-holders' works. A Suno-Musixmatch pipeline means every Suno output will be checked against a commercial rights database before it ships, which is a meaningfully different posture from "we added a watermark and called it disclosure."
Suno's bet is that labels, courts, and regulators will accept a self-imposed regime if the regime is technically robust and arrives first. That bet has structural problems. Watermarks do not retroactively license a model trained on unlicensed music. They do not compensate the artists whose work was used to build the model. And they do not give the labels anything they would accept as a substitute for a binding training-data consent regime.
The technology choice Suno makes in the next few months will determine whether this announcement buys the company a regulatory off-ramp or just buys a year before a label- or court-imposed standard lands instead. The harder question, whether AI-music platforms can train on copyrighted catalogs without explicit permission, is the one the watermark does not answer at all.