Sony just named names, all 30,000 of them, and suddenly AI music training isn't an abstract legal theory anymore.
The Summary
- Sony Music filed a new lawsuit against Udio listing over 30,000 specific songs, from Elvis Presley's "Hound Dog" to Beyoncé's "Say My Name," that allegedly trained the AI music generator without permission
- Sony gained access to Udio's training data through legal discovery from their 2024 lawsuit, and this 30,000-song list is "only a small portion" of claimed infringements
- The case transforms AI copyright fights from philosophical debates about fair use into concrete battles over specific works with specific rightsholders
The Signal
Most AI copyright lawsuits operate in the fog of "we think you trained on our stuff." Sony just turned on the floodlights. After months of legal discovery, they pulled receipts. Not vague categories. Not representative samples. Thirty thousand individual songs, each one a discrete copyright claim, each one owned by artists and labels who now have a name to point at in court.
This specificity matters because it flips the burden of proof. Before, AI companies could say "prove it." Now Sony is saying "here's the list, explain why each one was fair use." That's a different game. The original 2024 lawsuit with Universal and Warner was a warning shot. This one is a siege engine.
"Sony claims the 30,000 songs represent only a small portion of infringed works."
The economics here get wild fast. If statutory damages apply, the range is $750 to $30,000 per work, or up to $150,000 if willful. Do the math on 30,000 songs at the low end and you're at $22.5 million. At the high end, you're looking at $4.5 billion. Even if Sony only wins on a fraction of claims, the numbers get big enough to matter for a startup.
But the real stakes aren't damages. It's precedent. Every AI music company, every text-to-audio model, every synthetic voice platform is watching this. If Sony wins, it means you can't build a competitive music AI without licensing deals. That doesn't kill the technology. It just means the only companies that can afford to build it are big enough to cut deals with Sony, Universal, and Warner. The agent economy for music creation doesn't disappear. It just gets gatekept.
Key questions this raises:
- How many other songs are in Udio's training set that Sony doesn't own?
- What did the discovery process actually reveal about how Udio sources and processes training data?
- Can smaller AI music startups survive if they have to pre-license catalogs before launch?
The Implication
If you're building anything that generates creative output, watch how this case defines "transformative use" for AI training. The legal standard that emerges here will ripple across every generative model, from code to video to voice. The winners will be companies that either have deep enough pockets to fight multi-year legal battles or smart enough to structure licensing deals before they train.
For musicians and creators, this is the first real test of whether copyright law can actually protect you in the agent economy, or whether your work just becomes training data with extra steps.