German Court Rules ChatGPT Violated Law by Learning from Song Lyrics
London-UK, November 12, 2025
In a landmark legal ruling with global implications for the development of Artificial Intelligence, a regional court in Hamburg, Germany, has determined that ChatGPT and its underlying large language models (LLMs) violated copyright law when its training data incorporated protected material, specifically song lyrics, without the explicit permission of the rights holders.
The verdict is a major victory for creative industries and sets a formidable legal precedent in Europe, asserting that the act of “learning” or copying expressive works during the mass-scale data processing known as text and data mining (TDM) is not shielded by existing fair use or European intellectual property exceptions.
The ruling could fundamentally reshape how AI companies must secure and document their massive training datasets, forcing them to negotiate licensing deals for every protected work they consume.
Key Headlines
Landmark Ruling:
The Hamburg Regional Court ruled in favour of two major German music publishers and a songwriters’ collective, stating that the use of protected song lyrics constitutes a copyright breach.
TDM Exception Rejected:
The court rejected OpenAI’s argument that its use of copyrighted works fell under the European Union’s Text and Data Mining (TDM) exception, which allows for non-commercial research use.
Licensing Requirement:
The ruling suggests that for commercial, profit-driven LLMs like ChatGPT, developers must obtain prior licenses for all copyrighted materials, or face severe financial penalties.
Global Implications:
The decision is expected to embolden artists and copyright holders across the EU and the UK to pursue similar actions, potentially forcing a significant overhaul of AI training practices worldwide.
The case was brought forward by two prominent German music publishing houses and a collective representing thousands of local songwriters.
The core of their argument focused on the training process of OpenAI’s generative AI models. They presented evidence demonstrating that when prompted, ChatGPT could accurately reproduce and generate derivative works based on specific, copyrighted song lyrics whose use had never been licensed to the AI firm.
The plaintiffs argued that the copying of these works—an essential step in the massive data mining used to train the LLMs—constituted an unauthorized reproduction and distribution of their protected material.
The Hamburg court’s key determination hinged on its interpretation of the European Union’s Copyright Directive, specifically the provisions related to Text and Data Mining (TDM).
While the EU directive does provide some exceptions allowing for the use of copyrighted material for TDM, the court ruled that these exceptions were not applicable to OpenAI’s commercial, large-scale training of its for-profit LLMs.
The judgment effectively stated that since the primary purpose of the resulting LLM is commercial gain, the entire training process is inextricably linked to commercial exploitation, thereby demanding proper licensing.
This legal stance directly contrasts with some interpretations of “fair use” in the United States, where the use of copyrighted material for transformative purposes, such as AI training, has seen mixed results in court.
The German ruling firmly establishes that in the EU legal environment, the use of protected content for commercial LLM training requires a proactive and specific license from the rights holder.
The ruling sends a clear message: AI companies cannot simply scrape the internet, including copyrighted works, and claim an implicit right to use them to build multi-billion-dollar products.
The potential ramifications of this decision are immense.
Firstly, it opens the door to a deluge of similar lawsuits from musicians, authors, visual artists, and news publishers across Europe, all of whom can now point to this precedent to demand compensation for the unauthorized use of their creative output.
Secondly, it creates a massive logistical and financial challenge for OpenAI and other LLM developers, such as Google and Meta.
They may now be required to conduct costly audits of their existing training datasets—which contain trillions of data points—to identify and retroactively license every single copyrighted work used, or risk substantial penalties and injunctions.
Should the ruling be upheld on appeal, the legal framework for AI development in the EU could mandate an “opt-in” system, where copyright holders must explicitly agree to their work being used for TDM, rather than the current implicit “opt-out” system.
This shift would significantly increase the cost and complexity of training new models, fundamentally changing the economics of the generative AI industry in Europe and accelerating a global push toward creating fully licensed, transparently trained AI models.
