A Munich court ruling issued on 11 November 2025 found that OpenAI's GPT-4 and GPT-4o models infringed German copyright law by memorising song lyrics, and a further reported decision from the same court suggests that training on protected content may engage German copyright liability even when that training occurs outside the European Union. These decisions, taken together with ongoing proceedings at the Court of Justice of the EU, represent a material shift in the legal risk profile for any international company engaged in AI model development or deployment that touches German-protected works. Businesses should treat the current period as an active compliance window rather than a period of regulatory stability.
What the Munich Court Decided and When
On 11 November 2025, Landgericht München I (case reference 42 O 14139/24) ruled in GEMA v OpenAI that the memorisation of song lyrics by GPT-4 and GPT-4o constituted copyright infringement under paragraphs 15, 16, and 19a of the Urheberrechtsgesetz (UrhG). The court held that the text and data mining (TDM) exception under paragraph 44b UrhG did not apply on the facts presented. OpenAI is appealing the decision, meaning it is not yet final, but it carries significant weight as the first substantive German ruling directly addressing large language model outputs and copyright infringement.
Of equal or greater practical significance is a further LG München I ruling reported on 31 July 2026. That decision – which is also not final – indicated that training on protected musical works may constitute infringement under German law even where the training process took place outside the EU. This extraterritorial framing, if upheld on appeal and ultimately confirmed by the Bundesgerichtshof, would substantially expand the jurisdictional reach of German copyright law over AI developers headquartered in the United States, United Kingdom, Canada, or elsewhere.
Companies with AI products distributed or accessible in Germany should treat both rulings as live risk indicators, even before any final determination is reached.
How This Diverges from English Jurisprudence
The contrast with the English High Court's approach is instructive. In Getty Images v Stability AI [2025] EWHC 2863 (Ch), handed down on 4 November 2025, the court found that model weights do not store copies of protected works in a legally cognisant sense, and the secondary infringement claim consequently failed. Trade mark findings in that case were limited in scope. The Munich approach – focusing on memorisation and output reproduction – proceeds on a different analytical path, one that may be more receptive to claimants asserting rights in expressive works such as lyrics, poems, and literary text.
This divergence matters for international companies that have modelled their compliance frameworks on English law or on the assumption of a harmonised European approach. Germany and England are now articulating distinct liability theories, and a product cleared in one jurisdiction may face real exposure in the other.
The EU Dimension: CJEU Proceedings in C-250/25
The Court of Justice of the European Union is also engaged on closely related questions. In Like Company v Google (C-250/25), a hearing took place on 10 March 2026 and an Advocate General opinion is expected in September 2026. That opinion, when published, will offer the first authoritative EU-level signal on how the TDM exception and related copyright provisions interact with AI training at scale. It will not be binding, but AG opinions carry considerable persuasive weight with the CJEU and with national courts – including German ones.
Until the AG opinion is published and the CJEU issues its judgment, national courts within the EU will continue to develop their own positions. Germany's courts have shown a willingness to move independently and at speed.
Who Is Affected in Germany
The following categories of business face the most direct exposure in light of these developments:
- AI model developers – whether incorporated as a GmbH or another entity – that have trained on text, lyrics, musical works, or other expressive content that enjoys German copyright protection, regardless of where training occurred.
- Platform operators and API resellers distributing AI-generated content to users in Germany, who may face secondary liability arguments depending on how the appeal courts develop the reasoning in GEMA v OpenAI.
- Rights-holding organisations and individual creators registered in the Handelsregister or otherwise domiciled in Germany, for whom these rulings open new litigation pathways.
- Foreign companies without a German establishment that serve German users or whose models were trained on German-protected content – the reported extraterritorial finding is directly relevant to this group.
- Investors and funds with portfolio exposure to AI companies, for whom copyright liability at scale is a material valuation and due diligence issue under German IP litigation risk analysis.
The Text and Data Mining Exception: Why It Did Not Assist OpenAI
Paragraph 44b UrhG implements the EU's TDM exception from Directive 2019/790. It permits text and data mining of lawfully accessed works for the purposes of scientific research and, under certain conditions, commercial use – but rights holders may opt out for commercial TDM by making a machine-readable reservation. The Munich court found that the exception did not apply on the facts of GEMA v OpenAI, although the precise reasoning will remain subject to appellate scrutiny.
This outcome signals that the TDM exception cannot be assumed to function as a general safe harbour for large-scale AI training in Germany. Businesses that have relied on it without conducting jurisdiction-specific legal analysis should revisit their position. Design registration and trade mark considerations may also arise where AI-generated outputs reproduce stylistic or protected elements of registered designs or marks.
Immediate Action Items for International Companies
Given the current state of proceedings – two non-final but significant Munich rulings, a pending CJEU opinion, and a divergent English judgment – the following steps are advisable in the near term:
- Audit training data provenance: identify whether any corpus used in model training included German-protected literary, musical, or artistic works, and whether any opt-out reservation under paragraph 44b UrhG was in place at the time of ingestion.
- Assess output memorisation risk: conduct technical analysis of whether deployed models reproduce protected content verbatim or in substantially similar form, as this was the operative finding in GEMA v OpenAI.
- Map jurisdictional exposure: do not assume that training conducted outside Germany or the EU insulates your company from German copyright claims; the reported 31 July 2026 ruling, if confirmed, would eliminate that assumption.
- Review licensing arrangements: consider whether existing or prospective licences with rights-holding organisations cover AI training and generation use cases under German law, and whether those licences need to be renegotiated.
- Monitor appellate and CJEU developments: the OpenAI appeal before the OLG München and the forthcoming AG opinion in C-250/25 are the two most consequential near-term developments; build a monitoring protocol into your compliance calendar.
For companies that are already subject to AI Act obligations in Germany, copyright compliance and AI Act compliance are increasingly intertwined. Our analysis of the intersection of these frameworks is available at AI Act compliance in Germany.
Litigation and Enforcement Outlook
The GEMA v OpenAI litigation is proceeding through the German courts at a notable pace. Germany has well-developed IP litigation infrastructure, and rights-holding organisations such as GEMA have the institutional standing and resources to pursue multi-round proceedings up to the Bundesgerichtshof if necessary. The Mahnverfahren (summary payment procedure) is not the relevant procedural pathway here; these are substantive IP claims litigated on full merits.
For foreign defendants, service of process, jurisdictional challenges, and enforcement of any eventual judgment are all procedural questions that require early attention. The reported extraterritorial finding heightens the urgency for non-EU companies that may have previously considered German proceedings a remote prospect.
If your organisation is already facing or anticipates German copyright claims in the AI context, early legal analysis is essential. Our IP and technology litigation team in Germany can advise on procedural options and strategic positioning. You are welcome to reach out via our contacts page or at info@ferrazwhitmore.com.
A detailed technical and legal analysis of the liability framework established in GEMA v OpenAI, including commentary on the TDM exception, memorisation doctrine, and extraterritoriality, is available in our deep-dive briefing at GEMA v OpenAI – Training Liability in Germany.
What to Watch Next
Three developments will most significantly shape the legal landscape over the coming months. First, the OLG München appeal in GEMA v OpenAI will either affirm, modify, or overturn the findings on TDM and memorisation. Second, the AG opinion in CJEU C-250/25, expected in September 2026, will signal the EU-level direction on AI training and copyright. Third, any further rulings from German courts on the extraterritoriality question – building on the reported 31 July 2026 decision – will clarify whether offshore training genuinely engages German copyright liability or whether that principle remains contested.
Companies operating in or distributing AI products into Germany should not wait for final judgments before updating their risk assessments. The direction of travel is clear, even if the precise legal boundaries are still being drawn.
Ferraz & Whitmore advises international clients on German and cross-border IP matters, including copyright infringement disputes, AI governance, and IP litigation in Germany. If you would like to discuss how these developments apply to your specific situation, please contact us at info@ferrazwhitmore.com or visit our contacts page.
Disclaimer: This publication is provided for informational purposes only and does not constitute legal advice. The information herein should not be relied upon as a substitute for professional legal counsel tailored to your specific circumstances. Ferraz & Whitmore assumes no liability for actions taken or not taken based on the contents of this material. For advice regarding your particular situation, please contact info@ferrazwhitmore.com.