Suno thought signing licensing deals with Warner Music Group and BMG would change the conversation. Apparently not. Sony Music Entertainment and Universal Music Group are now suing the AI music startup over its newest v6 models, arguing, as reported by Engadget, that the company found a roundabout way to bake unlicensed material into its latest product anyway.
The core of the argument is this: Suno publicly acknowledged that v6 was trained partly on users’ interactions with its service. Sony and UMG read that as a polite way of saying the model learned from outputs and preference signals generated by older models, which were themselves built on recordings the labels say were copied without permission. If that interpretation holds up in court, it doesn’t matter that Suno secured licensing agreements for v6. The contamination, as the labels see it, runs deeper than the training data itself.
The financial exposure here is significant. Sony and UMG claim Suno could be liable for infringing at least 60,202 sound recordings, which puts potential damages at up to $9 billion under US copyright law. On top of that, the lawsuit adds up to $2,500 per instance for allegedly scraping songs from YouTube by circumventing its anti-downloading technology. That last detail connects to a July 2026 data breach that exposed how Suno pulled millions of songs and lyrics from platforms including YouTube Music, Deezer, and Genius to train its earlier models.
Suno pushed back predictably. The company called the claims “fundamentally flawed on both the facts and the law” and pointed to its partnerships with WMG, BMG, and Believe as evidence of good faith. But the statement doesn’t really address the specific legal theory the labels are advancing, which is about derivative contamination, not just whether a licensing contract exists.
This matters beyond Suno. The entire AI music space, including competitors like Udio, which faced its own copyright litigation, is watching how courts define the boundaries of what counts as infringement when outputs from unlicensed models feed into licensed ones. If judges accept the labels’ framing, it creates a serious problem for any AI company that trained early models on scraped data and then tried to clean up its act later. Licensing deals going forward wouldn’t necessarily protect against liability from the past.
Suno also quietly shut down its older models alongside the v6 launch, which at least signals the company knows those earlier products are a legal liability. But sunsetting a model doesn’t erase the training decisions that built it. That’s exactly what this lawsuit is testing.
