AI Companies Stole Your Favorite Songs. Here's How to Check

Every time you ask ChatGPT to write a song or use an AI music generator, you're using a system trained on stolen music. The Atlantic just proved it.

Reporter Alex Reisner built a searchable database exposing four datasets that AI companies used to train their models. We're talking 21 million songs total. Taylor Swift, The Beatles, Radiohead, Kendrick Lamar — they're all in there. Without permission. Without payment.

The companies behind these datasets didn't ask. They just took.

What Actually Happened

Reisner uncovered four massive collections of copyrighted music that AI companies have been using as training data:

  • Dataset 1: 12 million tracks
  • Dataset 2: 9 million tracks
  • Dataset 3 & 4: Smaller collections, but still hundreds of thousands of songs

These aren't bootleg recordings or covers. These are the actual studio recordings you stream on Spotify. The ones artists spent years creating and own the rights to.

The datasets came from academic researchers and AI companies who scraped music from across the internet. They justified it as "research" or "fair use." But when your research powers billion-dollar AI products, that excuse falls apart.

You can search the database yourself at [The Atlantic's website]. Type in any artist name. The results will shock you.

Why This Matters for Everyone

This isn't just about rich musicians losing royalties. This affects every artist trying to make a living.

When AI can instantly generate music "in the style of" any artist, why would streaming services pay for the real thing? Why would brands license actual songs for commercials when AI can make knockoffs for free?

Smaller artists get hit hardest. That indie band you discovered on TikTok? Their music is probably in these datasets too. But they don't have Taylor Swift's legal team to fight back.

The precedent here is dangerous. If AI companies can take copyrighted music without permission, what stops them from taking your photos, videos, or writing? Some already have.

We're watching the largest copyright theft in history happen in real time. And it's being called "innovation."

The Industry's Weak Response

Record labels are finally waking up. Universal Music Group and Sony have filed lawsuits against AI companies. But the damage is done. The models are already trained.

Some AI companies claim they'll start paying for music licenses. Too little, too late. They built their entire business on stolen content first, then offered to pay after getting caught.

The legal system moves slowly. These cases will take years to resolve. Meanwhile, AI music generators keep pumping out content based on stolen training data.

Musicians are organizing, but they're fighting billion-dollar tech companies with infinite legal budgets. The power imbalance is staggering.

What You Can Do Right Now

Check if your favorite artists were stolen from: Search The Atlantic's database. See which of your favorite musicians had their work taken without permission. Share what you find. Public pressure works.

Support artists directly: Buy music from artist websites, not just streaming services. Go to concerts. Buy merchandise. When AI can copy their sound, your direct support matters more than ever.

Be skeptical of AI music: That perfect song an AI generated? It's probably a mashup of stolen elements from real artists. Credit the humans who actually created the building blocks.

The music industry survived file sharing, streaming, and social media. But AI theft is different. It's not just copying songs — it's copying the ability to make songs.

Artists spent decades fighting for fair streaming royalties. Now they have to fight for the right to own their own creative DNA.

— Dolce