top of page

Two Good Weeks for Human Music — and the One Thing Neither of Them Fixed

14. Aug.
7 Min. Lesezeit

The Human Frequency — Notes on making music by hand in the age of AI

By Roger Ricks


On 31 July 2026, the Regional Court in Munich ruled that Suno, probably the best known AI music generator in the world, infringed copyright by training on songs it had no licence for. Eleven days later, on 11 August, Spotify announced it will start putting an "AI Persona" badge on artist profiles that turn out not to be people.

If you've read anything I've written here, you'd expect me to be delighted. I'm not, quite. Both of these are real progress, and I want to give them full credit before I tell you what still bothers me.


What the court actually said, in plain language


This is the part worth understanding, because most of the coverage stopped at "Suno lost."

Suno's defence was roughly this: training is not copying. The machine reads a lot of music, learns the patterns of it, and produces something new, the way a human learns by listening. Germany even has a text and data mining exception in copyright law that would allow something like that.

The court didn't buy it, for two reasons that are more interesting than the verdict itself.

First, the exception only applies if you got at the material lawfully. Suno had already conceded that it obtained training data by stream ripping, pulling audio off streaming services. Circumvent the lock, lose the exception. Second, and this is the one that will echo: the court found that the six disputed songs are stored inside the model. Not learned in the abstract. Reproduced. Give Suno a title and a style and out came something recognisably "Forever Young", "Atemlos", "Mambo No. 5", "Rasputin", "Big in Japan", "Daddy Cool". The judges read that as memorisation rather than general pattern learning, and storage is, in copyright terms, a copy.

Then Munich did something bolder. Suno trains in Massachusetts. The court took jurisdiction anyway, on the reasoning that the outputs are served to users in the EU. Case number 42 O 763/25, if you want to look it up. GEMA's CEO Tobias Holzmüller put it in one sentence: AI models built on stolen intellectual property have no protection under the law.


Now the fair version

Here's where I have to be honest about the size of this.

The ruling covers six songs. Six. Suno has to stop reproducing and training on those specific works and disclose what it earned from them, which is how the damages get calculated. It's a first instance judgment, it's appealable, and Suno has said it disagrees and is weighing an appeal. The jurisdiction reasoning, which is the genuinely radical part, hasn't been tested by a higher court yet.

And Suno is not a company that six songs will hurt. It's valued at around $5.4 billion. In late 2025 Warner Music settled its own lawsuit against Suno and signed a licensing deal instead, complete with artist opt-in for names and voices and a paywall on downloads. Universal did something comparable with Udio.

Sit with the shape of that for a second. Launch the model, train on everything, get sued, settle, licence, carry on, now with a partner. The infringement becomes a line item. If you built the thing before the rules existed, the fine turns into a late entry fee for a market you already own.

So no, I don't think a German court has stopped anything. What it has done is establish a price and a principle, and the principle is the valuable part. Training data is not free just because it was reachable.


The badge that doesn't say what you think it says


Spotify's announcement is the one I've been waiting years for, and I read it twice because the first time I got it wrong.

From mid September 2026, profiles whose public identity (the name, the photo, the artist you think you're following) appears to be AI generated will get an "AI Persona" badge. Artists can disclose it themselves, Spotify's review team can apply it independently, and you can appeal. Badged profiles are excluded by default from every editorial and algorithmic recommendation. You can still find them. They just won't find you.

That's a real measure with real teeth. Losing algorithmic distribution is the closest thing streaming has to a penalty. Deezer has been doing something similar for a while, and it's why fully AI generated tracks account for only about 1 to 3% of actual listening there, despite making up more than half of daily uploads. Around 90,000 fully AI generated tracks a day at the June 2026 peak. The flood is enormous. The listening is not. Recommendation exclusion is most of the reason why.

But read Spotify's own sentence. This badge is about the artist's public identity, not about how the music was made.

So a fabricated person with a fabricated face gets flagged. A real human being who types a prompt, gets a finished track back and uploads it under their own name does not. The badge catches the deception about who. It says nothing at all about the what.

Spotify does have a separate answer for that. There are AI credits in the song details, built on the DDEX standard and announced on 25 September 2025, and there's SongDNA, which rolled out in March 2026 to show the web of people behind a track. I like both of them. But the credits are self disclosed. The people with the strongest reason to lie are exactly the people the system asks politely.


Why this lands on me specifically

I released a single called "That Day" in July and I left the mistakes in it on purpose. A guitar that fades a hair early. Backing vocals that come in a tick before they really should. I sing it with a Swiss accent I have no intention of fixing. Every note on it was written, played and sung by a person, and it was mixed in a room where I also build the guitars.

None of that is visible on a streaming page. My track and a prompt generated one sit on the same shelf, in the same font, with the same play button. And if the person behind the prompt used their own real name and face, nothing in either of these two announcements tells you which is which. Not the court ruling. Not the badge.


That's the gap. Both of these interventions are aimed at fraud: stolen training data, fake identities, bot farmed streams. Deezer found that up to 85% of streams on fully AI generated tracks were fraudulent in 2025. Fraud is the easy case. Everybody agrees fraud is bad. Nobody has yet had to make the harder decision, which is whether a listener has any right to know how the thing they're listening to was made.


Where I land, and what would change my mind

I think the badge belongs on the track, not on the profile. And I think the disclosure has to be a positive claim rather than an accusation. Not "this might be AI" but "this was played by people", declared by the artist, tied to a credit list, and enforced the way false credits get enforced everywhere else in this industry. Put it next to the writer and producer credits, where it has belonged since the first day this was possible.

Let me tell you what I'm watching for, so you can hold me to it. Whether a higher court upholds the Munich jurisdiction reasoning, because without that the ruling is six songs and a headline. Whether GEMA turns the win into an actual licence with actual per use payouts, or whether it ends up looking like Warner's deal after all. And whether Spotify extends the badge from identity to method within the next year, because if they can build detection good enough to spot a generated face, the argument that they can't do it for audio starts to get thin.

Change any one of those and I'll say so here.

The ground is moving fast and I could be wrong about parts of this. So I'll leave it as a question I'd actually like the answer to:


If Spotify put a small line under every track saying who played what, would you look at it? Or would you rather not know?


Tell me honestly. And if you want to hear what one stubborn human still does by hand, my music lives here.


Sources & further reading

1 Kommentar

Sam Steiner
Sam Steiner
15. Aug.

I definitely want to know that the music I listen to is not generated by AI. The problem is, now I have to check their profile, see if they have live gigs coming up (easiest way to see they are real), check out how many songs they released since AI is here and stuff like that... absolute pain in the ass. Which is bad for new, emerging artists - because the easiest way to know the music is not AI generated is to listen to band you already know, or the "big ones" who have already "made it". I like to support small artists, though. Sam (Tipino)

Gefällt mir
bottom of page