GEMA Beats Suno: What the 'Memorization' Ruling Actually Means for Campaign Music
By Marius Rieg · · 3 min read
Summary: Munich's Landgericht München I ruled in GEMA's favor against Suno – the model infringes copyright on six well-known songs, including "Atemlos durch die Nacht" and "Rasputin," because it demonstrably "memorized" them. The ruling isn't final; an appeal is considered likely. A perspective on why the "memorization" standard is narrower than the headline "AI music loses in court" suggests, and what that means in practice for campaign music made with Suno or Udio.
On July 31, 2026, Munich's district court (Landgericht München I) ruled in the first major European case involving an AI music generator: Suno infringes copyright on six well-known songs from GEMA's repertoire – "Atemlos durch die Nacht," "Rasputin," "Big in Japan," "Forever Young," "Mambo No. 5," and "Daddy Cool." The court prohibited four specific acts: training in the US, reproduction through storage in the model in Germany, making the model publicly available, and reproduction through the generated outputs themselves. The ruling isn't final; an appeal is considered likely.
What the ruling actually decided
The headline "AI music loses in court" oversimplifies what was actually decided. This isn't about AI-generated music in general, but about six specifically named, world-famous songs where the court could demonstrate the model reproduces them accurately enough to count as reproduction – not inspiration, but a retrievable copy of protected material. A model that generates a completely new track in the style of Eurodance or German Schlager on a prompt, without reproducing a specific, identifiable work, sits in a different legal position than one demonstrably capable of outputting "Atemlos durch die Nacht" nearly note for note.
The actual standard: memorization
The court works with a concept that matters more for practice than the overall verdict: memorization – a model's demonstrable ability to reproduce a specific training work nearly unchanged on retrieval. That's the same underlying idea I described yesterday regarding Article 50's process boundary for CGI: there, too, the legal assessment doesn't hang on the outward result but on what can be proven about the process behind it. For images, the standard is whether an AI system was involved. For music, it's whether a specific protected work is stored in the model in a way that lets it be deliberately retrieved. In both cases, what counts isn't how the result looks or sounds, but what can be proven about how it came to exist.
What this means for campaign music
For the practical use of Suno or Udio in ad jingles or campaign music, this produces a clear but narrow rule: the risk concentrates on prompts that deliberately reference or aim to recreate a specific song, artist, or title – "in the style of" plus an artist's name is exactly the phrasing most likely to push a model toward outputting memorized material. A generic style prompt with no reference to a specific work carries, under the current ruling, meaningfully lower legal risk, because there's no demonstrable memorization of a particular song involved – just a new combination of trained patterns.
What's still open
Important for your own judgment: the ruling isn't final, and the expected appeal centers partly on extraterritorial reach – that training conducted in the US was assessed under German law – and partly on the exact evidentiary standard for memorization itself. Anyone producing campaign music with AI today should treat this as a directional guardrail, not settled law.
Conclusion
GEMA beat Suno, but not because AI music infringes copyright as a category – because six specific, well-known songs can demonstrably be retrieved from the model. For campaign music, that means the real risk factor isn't the tool, it's a prompt that deliberately conjures up a specific existing work. Generic style instructions remain the safer path – until the appeal is decided.