Munich Court: OpenAI Illegally Used Song Lyrics in ChatGPT Training; Damages Awarded, Appeal Possible

## The ruling at a glance Germany has drawn a clear line for AI training. On November 12, 2025, TechCrunch reported a landmark ruling in Munich. The court said ChatGPT illegally used protected song lyrics during training. Damages were awarded to GEMA, with an appeal still possible. At the Munich Regional Court (Landgericht München I), presiding judge Elke Schwager issued the decision. The court found OpenAI memorized protected lyrics inside its language models. It also found ChatGPT reproduced those lyrics in outputs. Each act infringed exploitation rights under German law, the court said. GEMA filed the case in November 2024. It represents roughly 100,000 composers, lyricists, and publishers in Germany. The suit focused on nine well-known songs. Examples included Herbert Grönemeyer’s “Männer” and Helene Fischer’s “Atemlos durch die Nacht.” OpenAI argued ChatGPT learns statistical patterns from large datasets. It said the system does not store or copy specific works. The company also argued outputs depend on user prompts. So users, not OpenAI, should bear liability. The court rejected those arguments. It clarified that internal memorization and external reproduction can both infringe rights. Damages were ordered, though the amount was not disclosed. OpenAI said it disagrees and is considering next steps. Public statements framed the dispute differently. OpenAI emphasized that the decision concerns a limited set of lyrics. GEMA called the verdict a precedent for Europe and a defense of creators’ livelihoods. Rights organizations view it as a milestone for licensing talks with AI developers. ## Why it matters for Morocco European courts are shaping rules for both inputs and outputs. Companies deploying generative AI must track training data provenance. They must also control outputs that risk reproducing protected works. This has practical consequences for Morocco’s AI ecosystem. Morocco’s AI scene is growing and pragmatic. Startups cluster in hubs like Technopark and university-linked incubators. Programs at institutions such as Mohammed VI Polytechnic University support deep‑tech ventures. Research centers like MAScIR drive applied innovation. Government actors already manage related policy areas. The Agence de Développement du Digital (ADD) coordinates digital transformation. The CNDP oversees personal data protection under Law 09‑08. The BMDA manages authors’ rights and licensing for creative works. The Munich ruling strengthens the case for licensing‑first AI. It suggests courts may treat training on protected works as exploitation. It also warns that reproducing memorized content can trigger liability. That guidance should inform Morocco’s startups and enterprises. ## Practical steps for Moroccan startups and enterprises Start with training data governance. - Create a source registry for all datasets, crawls, and synthetic data. - Prefer licensed corpora; contract with publishers and BMDA for music or lyrics. - Use Creative Commons content with compatible licenses; avoid CC BY‑NC for commercial products. - Filter out known copyrighted lyric repositories and song databases. - Deduplicate and limit near‑duplicate sequences that raise memorization risk. - Keep consent records; track license terms, territories, and expiration dates. - Implement opt‑out lists from rights holders where applicable. - Test models for verbatim memorization using exact‑match checks on risky categories. Control outputs to reduce infringement risk. - Add content filters to block lyric requests and long quotations. - Detect high n‑gram overlap with known lyric corpora and suppress outputs. - Cap max tokens to avoid long verbatim passages in completions. - Use retrieval‑augmented generation with licensed sources for factual recall. - Provide refusal messages that guide users to lawful sources or licensed services. - Log prompts that trigger refusals and improve filters through regular audits. Strengthen your vendor strategy. - Seek transparency reports from model vendors on training sources and exclusion practices. - Require contractual indemnity and copyright‑safe modes in the API. - Enable regional policies and stricter filters for EU users. - Maintain an internal escalation path for rights complaints and takedowns. - Offer a notice‑and‑takedown channel and respond quickly to rights holders. ## Policy and ecosystem moves for Morocco Policymakers can accelerate responsible AI while protecting creators. Clear guidance lowers uncertainty for startups and integrators. Industry bodies can coordinate licensing and norms. - BMDA could offer collective licensing options tailored to AI training and retrieval. - ADD could convene working groups on AI and copyright with industry and academia. - CNDP can update guidance where AI processing intersects personal data obligations. - Encourage open, well‑licensed datasets in Darija and Amazigh for language technologies. - Establish an AI sandbox with compliance guardrails and audit requirements. - Promote research on model memorization, evaluation methods, and output suppression. ## Sector snapshots: practical AI in Morocco Many AI uses in Morocco do not touch copyrighted lyrics. They can grow with minimal rights friction. Focus on applied value and measurable outcomes. - Agriculture: yield prediction, precision irrigation, and soil mapping. - Water utilities: leak detection, demand forecasting, and asset monitoring. - Logistics and mobility: route optimization, fleet maintenance, and demand planning. - Customer service: bilingual chatbots in Darija and French for call centers and e‑commerce. - Tourism: multilingual trip assistants, hotel service bots, and itinerary planning. - Fintech: fraud detection, risk scoring, and compliance monitoring. - Healthcare: appointment triage and imaging support under clinical oversight. - Education: tutor systems, exam prep tools, and language learning aids. ## Serving EU markets: practical compliance Moroccan companies often sell into Europe. The Munich decision signals increased scrutiny. You need EU‑ready processes and contracts. Treat compliance as product quality. - Separate EU deployments with stricter output filters and logging. - Use licensed sources for any retrieval that may include music or lyrics. - Document model versions, training sources, and release notes for audits. - Appoint a rights point of contact for European complaints. - Work with counsel on text‑and‑data‑mining exceptions and opt‑out mechanisms. ## FAQ‑style clarifications Does this ruling ban training on the open internet? It addresses training that memorizes and reproduces protected lyrics without permission. It does not settle all scenarios or content types. Outcomes can differ across jurisdictions. Could users alone be liable? The court said vendor liability can exist when models memorize and reproduce protected works. Responsibility may also depend on prompts and system settings. Seek counsel if your product enables risky outputs. Is fair use a defense here? The case applied German law, which differs from U.S. fair use. Legal standards vary across countries. Do not assume cross‑border equivalence. Are public lyrics pages free to use? Public availability does not equal a license. Copyright requires permission or a valid exception. Treat lyrics and music as high‑risk content. Will this apply in Morocco? The ruling is not binding in Morocco. It can still influence industry practices and licensing expectations. Align early with local rights holders and legal guidance. ## Key takeaways - The Munich court found ChatGPT memorized and reproduced protected lyrics, infringing German law. - Damages were awarded to GEMA; the amount was undisclosed, and appeal is possible. - Rights groups see a milestone for licensing and European AI governance. - Moroccan builders should adopt licensing‑first data pipelines and strict output controls. - Policymakers can pair creator protection with AI growth through clear, practical guidance.