What the court actually decided
On 31 July 2026, the 42nd Civil Chamber of the Landgericht München I ruled that Suno, the US-based AI music generator, infringed the copyrights of six well-known songs — including "Atemlos durch die Nacht," "Rasputin," and "Daddy Cool" — represented by the German collecting society GEMA (case no. 42 O 763/25). The court largely granted GEMA's claims for an injunction, disclosure of usage and revenue, and a declaration that damages are owed, with the exact sum to follow once Suno discloses its scale of use in Germany. The judgment is not final; an appeal to the Munich Court of Appeals is expected, mirroring the path GEMA's earlier win against OpenAI has already taken.
What makes this ruling notable isn't that a court found AI-generated outputs could infringe — courts across multiple jurisdictions have already gone there. It's the theory of where the infringement happened. GEMA's claim covered four distinct acts: reproduction of the works during training in the United States, reproduction through "memorisation" inside a model hosted on German servers, communication to the public by offering the model, and reproduction/communication through the actual outputs. The court found all four established. Critically, it reached the US-based training conduct by applying US fair use law itself — and concluded Suno's use failed that test, reasoning the outputs were substantially similar to the originals and could not be characterized as sufficiently transformative. It then held that the German-hosted memorisation and the outputs generated in Germany independently infringe German copyright law, regardless of the US analysis. The court also expressly held that compliance with the EU AI Act's training-data transparency obligations is not a defense to a copyright claim — the two regimes are cumulative, not substitutive.
The second Munich loss for a US AI company in nine months
This is not GEMA's first courtroom win against a generative AI provider. On 11 November 2025, the same Munich court ruled that OpenAI's memorisation and reproduction of nine German song lyrics through ChatGPT infringed copyright, ordering damages, disclosure, and an injunction — a decision OpenAI has appealed. The Suno ruling extends that reasoning from text to music and, notably, reaches conduct that took place entirely outside German territory. GEMA's general counsel framed the jurisdictional holding as the more consequential piece: European courts, the argument goes, can assert authority over AI systems that serve European users and store trained models on European infrastructure, regardless of where the underlying training computation ran.
The case for the ruling
The strongest argument for the court's approach is straightforward: if territorial location of a GPU cluster determines whether copyright law applies, every AI company has an obvious incentive to train exclusively in whichever jurisdiction has the loosest fair-use or text-and-data-mining rules, then simply serve outputs into markets with stricter protections. That would let a US company extract commercial value from a German songwriter's catalogue without ever triggering German law, so long as the actual GPUs sat in Virginia. Collecting societies exist precisely to prevent large-scale, systematic uncompensated use of a repertoire that individual songwriters could never police alone, and GEMA's position — that generative AI providers must license the same way streaming services and broadcasters do — is a coherent extension of decades of settled practice, not a novel imposition. A court unwilling to look past where the servers physically sit would be handing infringement a geography-shaped loophole.
Why the extraterritorial reach should still worry innovation-focused observers
But extending a national court's jurisdiction to conduct that occurred wholly abroad, assessed under a foreign country's own legal standard, is a significant expansion with consequences beyond this one case. Suno now faces the prospect of a German court's independent application of US fair use doctrine determining its liability for conduct that occurred on US soil — a role ordinarily reserved for US courts applying US law with the benefit of the full adversarial record and appellate correction mechanisms American defendants expect. If every jurisdiction where an AI model's outputs are eventually served claims the authority to adjudicate the lawfulness of training that happened elsewhere, AI developers face not one compliance regime but the union of all of them — the most restrictive rule anywhere becomes the effective global rule, since serving any market at all invites that market's courts to look backward at training. That outcome would chill smaller, well-resourced-but-not-Big-Tech developers far more than incumbents who can absorb multi-jurisdictional litigation risk, precisely the opposite of the innovation-diffusing effect proportionate copyright policy should aim for.
What comes next
The ruling is under appeal, so nothing here is settled German law yet, let alone EU law. The more consequential event is likely 3 September 2026, when the Advocate General is expected to deliver an opinion in Like Company v Google (C-250/25) — the CJEU's first case testing whether AI training and outputs infringe copyright under the EU's harmonized text-and-data-mining exception. That opinion, and the eventual Grand Chamber ruling, will bind all 27 member states and could either validate or narrow the Munich court's approach. Until then, GEMA has two wins and momentum; AI music and text providers serving European users have a live, unresolved extraterritoriality problem that licensing deals — not litigation — remain the fastest way to close.