Who Owns a Song Written by a Machine?

Imagine a song drops in the summer. It's catchy — the kind of thing that gets stuck in your head on a Tuesday afternoon when you're trying to focus. It racks up 40 million streams in three weeks. Then a journalist starts digging. Turns out an AI generated the chord progression, the melody, and 80% of the lyrics. A producer cleaned it up, tweaked a few words, and slapped a name on the release.

Now here's the uncomfortable question: who gets the royalty check?

That's not a hypothetical. It's the conversation happening right now in copyright offices, courtrooms, and music industry back rooms around the world. And the answer, frustratingly, is: it depends — and often, nobody.

Where the Law Currently Stands

The U.S. Copyright Office has been pretty clear about one thing: AI cannot hold a copyright. Their 2023 guidance, and subsequent rulings since, establish that copyright protection requires human authorship. A machine can't be an author. Full stop.

But that's actually the easy part. The harder question is what happens to the human in the loop.

The doctrine that's emerged is something called "sufficient human creative control." If a human made enough meaningful creative choices in the output, copyright can attach. The problem is that "sufficient" is doing an enormous amount of work in that sentence. Does selecting a genre and a tempo count? What about choosing from three AI-generated chorus options? What if you rewrote two lines of a 40-line lyric?

Nobody has a clean answer. Courts are working it out case by case, and the music industry is moving faster than the courts.

Three Scenarios, Three Very Different Outcomes

Think of AI involvement in music like a dial, not a switch. On one end, a songwriter uses a tool like Udio or Suno to explore melodic ideas the same way they'd noodle on a guitar — AI as instrument. On the other end, you push a button and a fully-formed, commercially-ready track comes out with no human creative input. Most of the interesting (and legally messy) territory lives in the middle.

The EU Is Moving Faster

While U.S. courts are still figuring out the "sufficient human creativity" threshold, the European Union's AI Act — which started phasing in during 2024 — has added a new layer: disclosure. If content was substantially generated by AI, it has to be labeled as such. That includes music.

What does this mean for a song released in Germany, France, or Spain? Platforms operating in the EU may soon be required to carry AI-origin metadata. That metadata gap between what's tagged and what's true is going to create legal exposure for labels and distributors who didn't think to ask.

Training Data: The Lawsuit Nobody Wants to Think About

There's another wrinkle here that the Getty Images vs. Stability AI case illuminated. It's not just about what AI produces — it's about what it was trained on. If an AI model learned to write country lyrics by processing a million copyrighted country songs without licensing them, the output may carry legal contamination from its origin, even if the resulting song sounds nothing like any individual source.

The music industry equivalent of the Getty case hasn't happened at scale yet. But it will. Several major labels have already filed or threatened suits, and the discovery phase of those cases will force AI companies to reveal exactly which catalog was used for training — and whether anyone got paid.

What This Means If You're a Rights Holder

You might be thinking: this doesn't affect me, I write all my own music. But here's the thing — it already might. Streaming platforms don't always know (or ask) what tools were used to create music. If someone releases AI-generated tracks that sound like your style, and those tracks collect royalties on the same PRO registrations, you have a problem. And if you've used AI tools and haven't documented how, your own copyright claims could be challenged.

Practically speaking, this means a few things:

Audit your catalog. If you've used AI tools — even just for ideas or inspiration — document what you used, how, and what creative decisions you made on top of it. "I gave it this prompt and used the output unchanged" is a very different story than "I used it to generate 10 chord options and built my own arrangement from scratch."

Tag your metadata. Rights metadata is already messy in this industry. Adding AI-origin information now — before it's mandatory — gives you a cleaner paper trail and, importantly, protects you in disputes. Platforms like Resolut let you attach that context directly to your assets, so it travels with the work.

Register what's registerable. If you have genuine human authorship, register it with your PRO and with the Copyright Office. A registration creates a timestamp. In a future dispute about whether something was human-created, that timestamp matters.

The Bigger Question

Here's what I keep coming back to: copyright law was designed to incentivize human creativity. The bargain was always — create something, get a limited monopoly on it, eventually it enters the public domain. That cycle made sense when humans were the only ones creating.

What happens to that bargain when creation becomes cheap, fast, and machine-driven? Who does the incentive run to? The answer isn't just a legal technicality — it shapes whether human musicians can make a living, whether creativity gets valued, whether the culture we make together means anything.

The Copyright Office doesn't have a full answer. The courts don't. The EU is trying. And meanwhile, the machines keep writing songs.