A German Court Just Ruled AI Can't Train on Music Without Permission. Here's How to Put Your Position on Record.

On 31 July 2026, the Munich Regional Court ruled that the AI music generator Suno infringed copyright by training its system on protected songs without a licence. The court found that the works were effectively memorised inside Suno's models and reproduced in its outputs, and it rejected the defences Suno relied on — both the European text-and-data-mining exception and the US fair-use argument. Suno was made subject to an injunction and ordered to disclose revenues linked to the infringements. GEMA, the collecting society that brought the case, called it a decision of global significance. Suno has said it disagrees and is weighing an appeal.

Whatever happens on appeal, the decision makes one thing concrete for every independent creator: your permission is not a formality. It is a right courts are now willing to enforce.

What the ruling actually establishes

Strip away the legal detail and the core finding is simple: an AI company cannot scrape catalogues of music, train on them, and sort out permissions later. The court treated the training inputs themselves as infringing uses — meaning the problem is not only what the AI produces, but the unlicensed material it was built from in the first place.

That shifts the burden. For a long time the practical default was that music was fair game unless someone actively fought back. This ruling points the other way: use without permission is the problem, and permission has to come first.

Why this matters even if you are not a GEMA member

You might not be represented by a European collecting society. You might be an independent artist releasing music yourself. The ruling still changes your world in one important way: it establishes that the creator's permission is the thing that counts — and that puts a spotlight on a question most independent music has never answered clearly.

Has the creator of a track actually stated whether it may be used for AI training?

For the overwhelming majority of independent releases, the answer is: nobody knows. There is no record. No stated position. And in a world where that permission now carries legal weight, silence is the weakest position to be in, because it leaves the question open for someone else to answer in their favour.

The gap this exposes

Here is the uncomfortable part. Owning your copyright — which, under the Berne Convention, you do automatically the moment you create a work — is not the same as having a clear, dated record of what you permit.

Copyright says the work is yours. It does not, on its own, create a dated statement that says this creator does not permit this recording to be used for AI training, made on a specific date, before any dispute arose. That kind of record is exactly what is useful when the question comes up — and after a ruling like this one, it will come up more, not less.

Right now, most creators have no such record at all.

How to put your position on record

This is what the AI-training declaration on an Audiverify certificate is for. When you document a track, you state your terms:

• Not allowed — the default. If you say nothing, your certificate records that AI training use is not permitted. The restrictive option is automatic, on purpose: permission should be something you actively give, never something assumed on your behalf.
• Allowed — if you want your work available for training (some creators do, especially for licensed or open datasets), you can say so explicitly.
• Allowed with attribution — permitted, but only with credit.

Whatever you choose is recorded word-for-word on your certificate and shown as your declared position, on a dated record created before any question was raised. To be precise about the mechanics: the independent RFC 3161 timestamp is applied to the SHA-256 fingerprint of your audio file, which is what fixes the file and its date. Your declaration is recorded on that same dated certificate alongside it — it is part of the record, not separately signed by the timestamp authority.

And to be clear about what this is and is not: it is a declaration recorded on your certificate, not a technical signal embedded in your audio file, and no such standard is widely honoured across the industry yet. Audiverify documents your stated position. It does not monitor or control how third parties use your file, and it cannot compel any AI company to honour your terms. What it does is remove the ambiguity: it turns "nobody knows what this creator wanted" into "the creator said no, on this date, verifiably."

The direction of travel

The Suno decision is one court, in one country, and it may be appealed. But it does not stand alone. The same Munich court ruled against OpenAI in a GEMA case in November 2025, finding that memorising song lyrics in model parameters was an unauthorised reproduction — and that ruling is not final either. Meanwhile the wider industry is visibly splitting between systems built on unlicensed material and licensed alternatives. The pressure is moving one way: toward permission, toward licensing, toward creators being asked what they allow.

If that is the direction, the creators who are ready are the ones who can point to a clear, dated record of their terms — not the ones scrambling to establish a position after the fact.

You made the work. Courts are increasingly agreed that your permission matters. This is one concrete, low-effort way to make that permission a matter of record, before anyone asks.

Related reading

• Can AI Train on Your Music? How to Declare That It Can't
• The Sony v. Suno Ruling Is Coming — What It Means for Independent Creators
• AI Music Copyright in 2026: Suno, Udio and the EU AI Act

Audiverify issues timestamped documentation for music releases, including AI-use and AI-training declarations, giving independent creators a clear, dated record of their stated terms. It documents what you declare — it does not grant rights, and it cannot control how others use your file. This article is general information, not legal advice.