Record labels accuse Suno of ripping YouTube tracks for AI training

The music industry's heavyweight legal machine is tightening the noose around generative AI.

On September 19th, the Recording Industry Association of America (RIAA)—representing heavyweights Universal, Sony, and Warner—filed an amended complaint against AI music darling Suno. The headline charge: Suno didn't just scrape music; it allegedly pirated songs straight off YouTube using illicit "stream ripping" to train its models.

Cracking YouTube's rolling cipher

According to the updated lawsuit, Suno employed dedicated code to bypass YouTube's "rolling cipher" encryption—a technical barrier built specifically to stop unauthorized downloads.

The labels claim this deliberate circumvention violated YouTube’s terms of service and ran directly afoul of Section 1201 of the Digital Millennium Copyright Act (DMCA), the infamous anti-circumvention rule.

While Section 1201 has historically been criticized for blocking everything from phone unlocking to fixing McDonald's ice cream machines, the RIAA is wielding it right at its original target: stopping piracy. And so far, no legal exception exists allowing AI companies to break digital locks just to grab training data.

The financial stakes are massive. The labels are seeking statutory damages of $2,500 per act of circumvention, on top of up to $150,000 for each copyrighted track infringed.

Demolishing the fair-use defense

Until now, Suno's primary shield has been the "fair use" doctrine—arguing that ingesting copyrighted media to teach a machine new skills is legally protected. It is an argument with some case law momentum, but zero industry consensus.

The RIAA's new angle, bolstered by research from the ICMP music publishers group, aims to gut that defense entirely. If Suno broke the law the moment it acquired the audio files, the fair-use defense largely evaporates. You cannot claim fair use on stolen property obtained through digital break-ins.

It is a legal playbook reminiscent of the high-stakes book piracy claims against Anthropic, and it signals a tactical shift across copyright holders: don't just argue about the output; prove the input was acquired through outright theft.

Why it matters

For creators and casual users, this lawsuit should be a massive yellow flag. Suno makes generating polished tracks absurdly easy, but if the courts rule that its foundational models were trained on illegally acquired data, the tool could be forced offline, stripped of its weights, or forced to pay back-breaking settlements that upend its business model.

For the broader AI industry, the RIAA's move is brilliantly tactical. Copyright infringement arguments around AI training are messy, philosophical, and slow. DMCA Section 1201 anti-circumvention violations, however, are cut-and-dried: either you bypassed the encryption, or you didn't.

We think this shifts the battlefield entirely. If breaking platform encryption to train AI models is established as an open-and-shut DMCA violation, the era of frictionless web scraping is officially dead. Startups won't be able to hide behind fair use if their data pipelines start with digital lock-picking.