Close-up of a Solid State Logic mixing desk with red indicators
Technology
💬 0

AI in music: what is already banned, what gets labelled and who is counting it

In a few weeks the industry took more decisions about AI than in the whole previous year: a Munich court, shared chart rules, a Beatport ban, a Spotify badge. We have put it all in one place — with dates, figures and links.

In spring the conversation about artificial intelligence in music was still an argument about taste: like it or not, "that is not art" versus "it is only a tool". Over the summer it stopped being an argument and turned into a list of decisions. A court set the line for lawful model training. Labels set the line for the charts. A store set the line for its catalogue. A streaming service set the line for an artist profile. And a detector finally showed how much of this there actually is.

Below is everything already decided, with dates, figures and a link to each individual story.

The court: training a model on someone else's catalogue without a licence is an infringement

Munich Regional Court ruled that Suno had no right to process works from the catalogue of GEMA, the German society of composers, lyricists and publishers. The court ordered the company to disclose revenue connected to the infringement and found it liable for damages, the amount of which is still to be determined. Suno disagrees and is considering an appeal. What exactly GEMA proved, and why it matters for generators.

The charts: "substantially created by a human"

Sony Music, Universal Music Group and Warner Music Group, together with Believe, BMG, Concord, Dirty Hit, Partisan Records and others, set out shared principles for tracks made with AI. The core requirement is that a recording be "substantially human-created". Two more follow: the music must raise no suspicion of chart manipulation, and any AI services used must be "properly authorised and lawful". The full wording, and why the labels talk about streaming fraud.

The store: Beatport no longer accepts fully generated tracks

The world's largest electronic music store has updated its upload rules: fully or mostly generated tracks do not enter the catalogue, while AI-assisted music is accepted but tagged during ingestion. Detection runs through Beatdapp. The store also cited a DJ survey: six in ten would not play a generated track in a set. All the survey figures and what Beatport's chief executive said.

Streaming: Spotify labels a generated artist persona

From mid-September profiles will carry an AI Persona badge, meaning the artist's name and face may be generated and may not represent a real person. Such profiles will drop out of recommendations by default. Artists can apply the badge themselves through the artist dashboard, but Spotify will review profiles on its own too, starting with those above certain audience thresholds. What both forms of the badge look like.

How much of it there actually is

The sharpest answer so far comes from SubmitHub's SH Labs detector, which checked more than a million July releases. 23.2% came back fully generated, and another 15.3% were machine audio a human had then processed. Together: almost 40% of everything released that month.

But the most uncomfortable figure is not about machines. Among artists whose releases tested positive this year, 31% answered a direct question by saying they had used no such tools at all. The full study, and who already uses the detector.

What the musicians themselves say

Thomas Bangalter — half of Daft Punk, the duo who spent twenty years performing in robot helmets — spoke about AI in creative work for the first time this summer. And said precisely the opposite of what was expected of him.

"The most interesting part of the creative process happens not at an intellectual level at all. It is a process in which you actually work out your own feelings," Bangalter said.

The whole conversation, and why the machine does not go there.

The questions people ask most

Is it allowed to make music with AI at all?

Yes. None of the decisions taken bans the tool. What they ban is something else: passing a fully generated track off as human, training a model on someone else's catalogue without a licence, and staying quiet about the machine's involvement. Beatport accepts AI-assisted music and the labels let it into the charts — provided the recording is "substantially human-created".

What does "substantially created by a human" mean?

There is no precise technical definition yet — the label coalition framed it as a principle, not a percentage threshold. In practice the platforms draw the line: Beatport filters out anything "fully or mostly" generated and tags the rest.

How can you tell a track was generated?

Automatic detectors are already running: SubmitHub's SH Labs is used by Bandcamp and Traxsource, and Beatport works with Beatdapp's technology. Listeners see the result as a label — on the track page in a store, or on an artist profile in Spotify.

Is it lawful to train a model on other people's music?

The German precedent says no, not without a licence. Munich Regional Court found that Suno had infringed copyright by using GEMA's catalogue. The ruling can be appealed and the company is pursuing that, so there is no final word yet.

What happens to artists who did not disclose their use of AI?

The mildest outcome is a label applied by the platform itself. Beyond that: disappearing from recommendations, as on Spotify, or an upload refused, as on Beatport, with the rightsholder notified. Staying quiet has stopped being free.

What all these decisions have in common is that none of them is about the quality of the music. They are about whether a person knows what they are listening to, and whether the author got what is theirs. The argument about taste has not gone anywhere — it is simply being had on top of rules that did not exist before.

Comments

Loading…

Jump Into The Music
ONE//FM
AI in music: what is banned and who checks it | ONE//FM