Context:
- Last year, German music rights collecting society Gesellschaft für musikalische Aufführungs- und mechanische Vervielfältigungsrechte (GEMA) brought a copyright infringement action against OpenAI over song lyrics (November 14, 2024 ai fray article) in the Landgericht München I (Munich I Regional Court; comparable to a U.S. federal district court though state-operated). At a subsequent press briefing, GEMA officials said their members (composers as well as lyricists) were as fascinated as they were frightened by Generative AI (November 19, 2024 ai fray article).
- This case does not involve musical compositions, which are equally copyrightable and at issue in a separate case GEMA brought against Suno earlier this year (January 21, 2025 ai fray article).
- Several dozen AI copyright infringement lawsuits are pending worldwide, most of them in the United States. On Thursday, a U.S. federal judge blessed a $1.5B+ settlement between a book authors’ class action (lead counsel: Susman Godfrey’s Justin Nelson) and Claude AI maker Anthropic (September 25, 2025 ai fray article).
What’s new: This morning, the court’s copyright-specialized 42nd Civil Chamber (Presiding Judge Dr. Elke Schwager with Judges Mattes and Hahn) commenced the one-day trial in GEMA v. OpenAI. The presiding judge left no doubt in her introductory outline of the issues in the case that the panel deems OpenAI liable for copyright infringement. She noted that the facts (which OpenAI addressed in apparently voluminous pleadings) were actually “not complicated” given that OpenAI does not dispute the use of the copyrighted works at issue in the training of its ChatGPT model. The judges are furthermore unpersuaded that users, not OpenAI, bear the responsibility for ChatGPT’s outputs.
Direct impact:
- Theoretically, OpenAI could still change the panel’s view, or the panel could have second thoughts between the trial and the decision, which will likely be scheduled for 4-8 weeks from now. But based on what we heard from two different sources present in the courtroom, there is no realistic chance that OpenAI can turn this around.
- The most likely next step is an injunction, but in order for OpenAI to comply with an injunction without having to shut down ChatGPT in Germany as a whole, a license deal may be the only realistic option. Otherwise OpenAI risks contempt-of-court sanctions if any further unauthorized reproduction occurs.
- It is theoretically possible (but doesn’t appear likely at this point) that the court refers certain questions of to the European Court of Justice (ECJ), where it could be addressed simultaneously with a different AI copyright matter (May 26, 2025 ai fray article). The presiding judge noted that both parties requested a preliminary reference to the ECJ, but appeared only lukewarm to the idea.
Wider ramifications: Munich has been the world’s number one patent injunction hotspot for several years (as we often discuss on ip fray, and which is the reason the 42nd Civil Chamber was created in 2020 so as to relieve the patent-specialized panels from copyright cases). The same court could now become the world’s number one AI copyright injunction hotspot if other right holders bring additional lawsuits there (which would almost certainly be referred to the 42nd Civil Chamber. The threat of German copyright injunctions could up the settlement pressure on those facing U.S. class actions and other enforcement actions.
GEMA asserted the copyrights and personality rights of select lyricists, which is why the formal value of the dispute (a construct for the purpose of determining court fees and the reimbursable part of litigation costs) is only €600K (approximately $700K). What GEMA seeks to achieve here is clarification and guidance for the purpose of enabling fruitful licensing negotiations. This differentiates the case from damages-focused lawsuits. Technically, an injunction would also come with a declaration that OpenAI is liable for infringement damages because GEMA so requested, but that is not the purpose. It is a safe assumption that both sides, above all OpenAI, have spent far more money on this litigation already than any realistic damages award over a select number of song lyrics (as opposed to GEMA’s entire portfolio) would amount to.
An opinion by the ECJ on key questions of EU copyright law would obviously provide ultimate clarification, but only the courts of final appeal of EU member states have a formal obligation to make such a reference, and even they do not have to do so if there is no serious question to be raised. In response to a question be GEMA’s counsel, the court said that it would reflect on the idea of a direct ECJ referral.
It appears that the Munich court considers this case a slam dunk for GEMA with the sole exception of a secondary (if not tertiary) claim involving personality rights (comparable to the allegations of misattribution that have been raised in some AI copyright cases in the U.S.). The court suggested the withdrawal of that side show, and its cost implications would likely be negligible. Also, that legal theory is rather specific to German law. German Urheberrecht (author’s law) focuses on authors’ control over the use of their mental creations, even to the extent that certain rights become inalienable, but is the closest instrument to copyright.
The court appears to be fully aware of the enormous relevance of this case, and after the introductory outline offered counsel a 30-minute break for consultations with their clients.
In the court’s preliminary (and unlikely to change) opinion, the ways in which OpenAI uses the song lyrics at issue embodies all elements of copyright infringement, from unauthorized use for training purposes all the way to unlawful reproduction.
OpenAI’s affirmative defense based on the text and data mining (TDM) exception under EU law has zero traction with the court. Even though the court recognizes that the TDM exception may inure to the benefit of AI technologies, the way OpenAI uses the copyrights-in-suit is not covered by it. For example, it does not amount to text and data mining for scientific purposes.
OpenAI does not do itself any favor with a throw-in-the-kitchen-sink approach, unselectively raising any defense in the book even if it amounts to an insult to human intelligence (such as alleging abusive litigation). From the perspective of a litigation watcher, I believe OpenAI would be better advised to approach such cases with humility.
Florian Mueller
Furthermore, I would urge OpenAI to think hard about whether it is a good idea to attempt to shift liability to users. Even if OpenAI believes that consumers would not be liable for copies made for private use, companies should strive to protect (and not hide behind) their customers. And there are professional ChatGPT users, too.
The court rejects OpenAI’s suggestion that users provoke copyright violations through their prompts. GEMA showed the court various prompts that resulted in a reproduction of copyright-protected text material. The court clarified that there could be future cases in which different prompts are on the table, but in this particular case, OpenAI cannot defend itself by shifting liability to users.
Any OpenAI theories that would be the next best thing under EU and German law to the U.S. fair use exception were rejected. For example, what ChatGPT does is not tantamount to citations or parodies.
On the bottom line, OpenAI is liable because of its central role and because it makes the copyrighted material publicly available.
OpenAI got short shrift for alleging an abusive nature of this lawsuit (which in other jurisdictions would be called “vexatious and oppressive”).
At this point it appears that OpenAI’s battle-of-matériel strategy (with seven lawyers from an extremely expensive firm that is not even specialized in copyright law, at least not in Germany, but primarily active in patent cases) has not paid dividends. GEMA sent only three lawyers to the trial.
OpenAI submitted a third pleading (after its answer to the complaint and sur-reply), which was received by the court in time for the trial, but served only today. Presumably OpenAI hopes that the court will identify anything in that pleading that will lead to a reopening of the record, i.e., a retrial. But that tactic rarely works.
After the break, GEMA’s lead counsel simply declared himself in agreement with the court. OpenAI’s lead counsel started to dive into highly technical aspects of OpenAI’s i inner workings, seeking to distinguish OpenAI’s use of copyright material from traditional and consistent copying. Of course, LLMs internally use statistical correlations, but those arguments did not sway the court in the build-up to today’s trial and appear unlikely to gain traction now. Based on what we heard from inside the courtroom, the panel was (as it must) paying attention, but did not appear to be dissuaded from its assessment of the case as of when this article was published.
OpenAI’s lead counsel used a tactic he has employed in some patent cases when he faced a hostile court: he tries to instill doubt in the judges that their decision would be appeal-proof, taking extreme positions on statutes and appellate case law.
The court will announce its decision in open court on November 11, 2025, and ai fray will report.
Court and counsel
Munich I Regional Court, 42nd Civil Chamber: Presiding Judge Dr. Elke Schwager with side judges Judge Mattes and Judge Hahn.
Counsel for GEMA: Raue’s Dr. Robert Heine, a specialized media and copyright attorney whose practice also includes AI-specific matters.
Counsel for OpenAI: Quinn Emanuel (lead counsel: Dr. Marcus Grosch).