Where to sue over AI in Europe: Munich, Hamburg, London or Luxembourg?

Where to sue over AI in Europe: Munich, Hamburg, London or Luxembourg?

by

In AI copyright litigation in Europe, the forum you choose can decide the case. Munich has twice held that works memorised in a model are unlawful reproductions, Hamburg has accepted that building a training dataset can fall within the text and data mining exception, London has held that a model trained abroad is not an “infringing copy” in the UK, and the Court of Justice of the EU in Luxembourg is about to give the first EU-wide guidance.

Update (October 2026): At the hearing on 3 September 2026 in Kneschke v LAION, the BGH indicated that points of EU law may need clarifying and that it could refer the case to the CJEU; no date for a decision was announced. beck-aktuell, 3 September 2026.

Key takeaways

  • Copyright is territorial: each court applies the law of the country for which protection is claimed, so where the acts happen (training, storage, outputs) shapes the claim.
  • Munich Regional Court I found infringement through memorisation in GEMA v OpenAI (November 2025) and GEMA v Suno (July 2026). Both are first-instance rulings.
  • The Hamburg Higher Regional Court applied the TDM exception to dataset creation in Kneschke v LAION; the Federal Court of Justice (BGH) hears the appeal on 3 September 2026.
  • The High Court in London rejected Getty’s secondary infringement claim against Stability AI; Getty has permission to appeal.
  • The CJEU cannot be approached directly, but its ruling in Like Company v Google (C-250/25) will bind courts in every Member State.

Why does the forum matter so much in AI copyright litigation in Europe?

Two sets of rules decide where you can sue and which law applies. Within the EU, the Brussels I bis Regulation lets you sue a defendant in its Member State of domicile (Article 4), in the courts for the place where the harmful event occurred or may occur (Article 7(2)), or join co-defendants where one is domiciled if the claims are closely connected (Article 8(1)). If the defendant is not domiciled in the EU, as with many US model developers, each country applies its own jurisdiction rules (Article 6).

The applicable law follows the Rome II Regulation: under Article 8(1), infringement is governed by the law of the country for which protection is claimed. In practice, training in the US, hosting in Ireland and outputs in Germany may be three different legal questions under three different laws.

Munich: memorisation and outputs

The 42nd Civil Chamber of the Munich Regional Court I, which specialises in copyright, has handed down the leading German decisions on what a model contains and produces.

  • GEMA v OpenAI (11 November 2025, 42 O 14139/24): nine song lyrics memorised in ChatGPT’s models were unlawfully reproduced; the TDM exception in section 44b of the German Copyright Act covers analysis, not storage of works (LTO, 11 November 2025). OpenAI’s appeal is pending before the Higher Regional Court of Munich as case 6 U 3662/25 e, according to the Hogan Lovells AI litigation tracker.
  • GEMA v Suno (31 July 2026, 42 O 763/25): six songs memorised in models stored on servers in Germany; outputs from simple prompts infringed; and, applying US law to training in the US, no fair use (court press release, 31 July 2026).

Munich suits claimants who can show that a model reproduces their works with simple prompts. But jurisdiction over the US training in Suno rested on section 131 of the Collecting Societies Act, a forum reserved for collecting societies.

Hamburg and the BGH: datasets and opt-outs

Kneschke v LAION concerns an earlier step: building a dataset of image-text pairs used to train image generators. According to the BGH press release of 20 May 2026, the Hamburg Regional Court (27 September 2024, 310 O 227/23) and the Hamburg Higher Regional Court (10 December 2025, 5 U 104/24) dismissed the photographer’s claim. The appeal court held that downloading images to compare their content with their descriptions was text and data mining under section 44b, that a reservation of rights for works online is only effective in machine-readable form, and that the non-profit defendant could also rely on the scientific research exception in section 60d. The BGH (case I ZR 281/25) set the hearing for 3 September 2026, as Morgan Lewis also noted on 25 June 2026.

London: territorial limits after Getty v Stability

In Getty Images v Stability AI [2025] EWHC 2863 (Ch), decided on 4 November 2025, Getty abandoned its claim over training because there was no evidence that Stable Diffusion was trained in the UK. Its remaining copyright claim, that importing the model was secondary infringement, failed: the judge held that a model that does not store the training works is not an “infringing copy”. Getty won only limited, historic trade mark infringement over watermarks. In the judge’s view, such a claim belonged where the model was trained, not in England. Permission to appeal on the secondary infringement point was granted in December 2025, as CMS reported on 21 January 2026.

London remains a major IP forum, but the UK is outside the EU’s Brussels I bis framework, so a UK judgment and an EU judgment are separate enforcement exercises.

Luxembourg: the CJEU and Like Company v Google

In private copyright disputes, the CJEU rules only on questions referred by national courts. In Case C-250/25, a Hungarian court asked in April 2025 whether a chatbot showing text from press publications is a communication to the public, whether training an LLM is a reproduction, whether the TDM exception in Article 4 of the DSM Directive covers it, and whether outputs prompted by users are reproductions by the provider. The hearing took place on 10 March 2026 and the Advocate General’s opinion is expected on 3 September 2026, according to the European IP Helpdesk (24 April 2026).

Comparing the main forums for AI copyright litigation

Forum Leading AI decision Best suited to Watch out for
Munich Regional Court I GEMA v OpenAI (2025); GEMA v Suno (2026) Memorisation and outputs reaching German users First instance; appeals pending or possible
Hamburg courts and BGH Kneschke v LAION (OLG Hamburg, 2025) Disputes over scraping, datasets and opt-outs TDM exception applied to dataset creation; BGH ruling pending
High Court, London Getty v Stability (2025) UK acts, trade mark and passing off claims Training abroad is outside UK copyright; appeal pending
CJEU, Luxembourg Like Company v Google (C-250/25, pending) Binding EU-wide interpretation Only via a national court reference; takes time

What this means for your business

  1. Map the acts: where the data was collected, where the model was trained and hosted, and where outputs reach users.
  2. Gather reproducible evidence: exact prompts, dates, outputs and, where possible, proof that your works were in the training data.
  3. Choose the claimant: a collecting society, a publisher or the author may each have different procedural options.
  4. Check your opt-outs: confirm that reservations of rights are machine-readable before relying on them.
  5. Sequence actions: a German claim, a UK claim and licensing talks can support each other if they are planned together.

For rights holders in Spain, Portugal and Latin America, the same logic applies: works exploited in Germany or the UK are protected there, wherever the author lives. Our team for AI and digital asset IP strategy can map these options before you commit to a forum.

Where companies get forum choice wrong

  • Suing where the company is, not where the acts are. A claim about training abroad may fail in a court that only applies its own territorial law, as Getty found in London.
  • Weak opt-outs. A reservation hidden in terms of use may not count as machine-readable.
  • Uncoordinated parallel actions. Inconsistent positions in different countries can be used against you; a single strategy across IP litigation and enforcement avoids that.

Frequently asked questions

Can I sue an AI company in Europe if it trained its model in the US?

Often yes, for acts in Europe: storing a model on European servers or generating infringing outputs for European users. Claims about the training itself depend on the forum. Munich applied US law to US training in GEMA v Suno, but on a jurisdiction rule for collecting societies; London declined a training claim where no training took place in the UK.

Which European court is most favourable to rights holders in AI cases?

No court guarantees a result. Munich has twice found infringement through memorisation, while the Hamburg courts applied the text and data mining exception to dataset creation. The right forum depends on the acts you can prove, where they occurred and who brings the claim.

When will the CJEU rule on AI training and copyright?

The Like Company v Google case (C-250/25) was heard on 10 March 2026, and the Advocate General’s opinion is scheduled for 3 September 2026, with the judgment to follow later. It will bind national courts on the questions referred, including reproduction in training and the TDM exception.

Can IP Global Guard coordinate AI litigation across several European countries?

Yes. We analyse the acts, evidence and options in each jurisdiction, coordinate qualified litigation counsel in Germany, the UK and elsewhere, and keep one strategy and one point of contact. Where the same rights are exploited in Latin America or Africa, we align those markets through local correspondents.

How IP Global Guard can help you choose the forum

AI cases are won or lost on early choices: which acts to target, which court to use and who should bring the claim. IP Global Guard, the IP services line of META Channel Corporation Limited, coordinates copyright, licensing and enforcement for creative and technology businesses across more than 25 jurisdictions in Europe, Latin America and Africa, and the group also advises on AI Act compliance.

Share the works concerned, the AI tools involved and the evidence you already hold. We will map the forums, compare the routes and set out a coordinated plan. Ask our team for a forum assessment.

This article is general information, not legal advice, and reflects the position on the date of publication.

Sources

Share

IP Global Guard

A specialist industrial and intellectual property service line of META Channel Corporation.

IP Global Guard is a service line of META Channel Corporation Limited, a Private Company Limited by Shares incorporated in Ireland. CRO 801240. Registered office: 34 Holywell Avenue, Feltrim Hall, Swords, Dublin, K67 KF22, Ireland.

© 2026 META Channel Corporation Limited. All rights reserved.