AI in court: GEMA wins verdict in Munich in historic trial against OpenAI's ChatGPT

More than just learned: Why ChatGPT's "memory" is now becoming a problem for OpenAI

A German court has delivered its verdict, and the echoes are reverberating from creative studios across Europe to the executive suites of Silicon Valley: In the landmark case of GEMA versus OpenAI, the Munich Regional Court ruled that ChatGPT infringed the copyrights of German musicians. At the heart of the proceedings were nine iconic German song lyrics, from Helene Fischer's "Atemlos" to Reinhard Mey's "Über den Wolken," which the chatbot could reproduce verbatim upon request. This ruling is far more than a legal victory for the approximately 100,000 artists represented by GEMA; it is a resounding victory in the struggle for the dignity and value of creative work in the age of artificial intelligence.

The conflict exposes the economic logic of a new digital expropriation: On one side are AI companies like OpenAI, which, with valuations in the hundreds of billions of dollars and rapidly growing revenues, create gigantic value. Their business model is largely based on a raw material for which they have not yet paid: the collective knowledge and creativity of humanity, which they use as training data. On the other side are artists, musicians, and authors who fear massive losses of income and the loss of their livelihoods due to AI-generated content.

The Munich ruling brings a key technical and legal question into focus: What exactly happens in the "brain" of an AI? While OpenAI argues that its models only learn abstract patterns, the court proves the existence of so-called "memorization"—the AI's ability to precisely store and reproduce copyrighted works. This undermines the arguments of the tech giants and opens the door to a fundamental renegotiation of the rules of the game. The Munich decision thus marks the beginning of a global debate that will define whether human creativity will continue to be fairly rewarded in the future or be reduced to free fuel for the next industrial revolution.

The battle for intellectual property in the age of artificial intelligence

When algorithms become free riders: The economic expropriation of the creative industries through generative AI systems

The ruling handed down by the Munich Regional Court on November 11, 2025, in the case of GEMA versus OpenAI marks a turning point in the debate surrounding the economic exploitation of creative work in the digital age. The decision in favor of the collecting society establishes that the operator of ChatGPT infringed copyright by using nine well-known German song lyrics. This is the first time in Europe that the highest court has confirmed what artists and rights holders have been arguing for years: the multi-billion-dollar technology companies of Silicon Valley are systematically appropriating creative work without compensating those whose labor forms the very raw material of their business models. However, this ruling is far more than a single legal decision. It reveals the fundamental tensions within an economic system in which the digital appropriation of human creativity has become the core mechanism of new accumulation strategies.

The economic dimensions of this conflict are considerable. OpenAI, which already generated $3.7 billion in revenue in 2024 and projects annualized revenue of $13 billion for 2025, builds its success on the free use of millions of copyrighted works. These works were used without permission or compensation to train the language model that is now used by over 700 million people weekly. The company's valuation reached an astronomical $500 billion in October 2025. This enormous value creation contrasts sharply with the increasing pressure on creative professionals: studies predict revenue losses of up to 27 percent for musicians due to AI-generated content, while the dubbing industry faces losses of up to 56 percent. The economic success of AI companies correlates directly with the anticipated decline of traditional creative professions.

The legal watershed and its background

The Munich ruling marks the end of a legal battle that began in November 2024 with the filing of a lawsuit by GEMA (the German performing rights society). At the heart of the case are nine song lyrics by prominent German artists, including Helene Fischer's "Atemlos," Herbert Grönemeyer's "Männer," Reinhard Mey's "Über den Wolken," and Rolf Zuckowski's "In der Weihnachtsbäckerei." GEMA, which represents approximately 100,000 musicians in Germany, was able to demonstrate that ChatGPT reproduced these lyrics exactly or almost identically in response to simple queries. This finding was considered proof that the lyrics were not merely used to train the model, but were stored or memorized within the system in a way that constituted ongoing reproduction.

The legal core of the proceedings revolves around the interpretation of the EU Directive on text and data mining, which was transposed into German law in 2021. Section 44b of the Copyright Act generally permits the automated analysis of works, provided they are lawfully accessible. This limitation was intended to promote innovation in the field of artificial intelligence without requiring developers to acquire licenses for each individual data set. However, paragraph three of the section stipulates that rights holders can object to such use. For works available online, this objection must be made in machine-readable form. GEMA had declared such an objection, the validity of which OpenAI contested.

The legal complexity lies in distinguishing between training a model and its subsequent use. While the Hamburg Regional Court ruled in September 2024, in a case concerning photographs, that creating training datasets could be permissible under certain conditions, the Munich court focused on the output of texts by ChatGPT. OpenAI argued that the model does not store data but merely reflects what it has learned from the entire training dataset. The output is generated through a sequential-analytical, iterative-probabilistic synthesis, not by retrieving stored content. GEMA, on the other hand, referred to technical studies showing that large language models can indeed memorize training data, especially if it occurs frequently in the dataset.

Judge Elke Schwager already indicated during the oral hearing in September 2025 that she was inclined to follow GEMA's arguments on virtually all key points. The now-announced verdict confirms this assessment and establishes that both the training with the protected works and their reproduction by the chatbot infringe copyright. The decision has no immediate binding legal consequences, as an appeal is expected. However, it sends a clear signal: In Europe, AI providers must acquire licenses if they wish to use copyrighted works.

The economic logic of digital appropriation

To grasp the implications of the Munich ruling, one must understand the economic mechanisms that have enabled the rise of the AI ​​giants. OpenAI operates within an economic structure that economist Philipp Staab has described as platform capitalism. Unlike classical industrial capitalism, where value creation primarily occurs through the transformation of physical goods, the platform economy is based on the control of data flows and access rights. Platforms like OpenAI create proprietary markets; they are the market itself. Their power is not based on the production of goods, but on the capitalization of resources that are, in fact, not scarce.

In the case of ChatGPT, this abundant resource is the freely available cultural and informational material on the internet. Through web crawling and systematic extraction of publicly accessible content, OpenAI and similar companies have amassed training datasets of a scale that defies all historical comprehension. The GPT-3 model was trained on approximately 560 gigabytes of text data, encompassing trillions of words. Acquiring this data was largely free, as the material was readily available online. However, the subsequent processing requires enormous investment: Training costs for GPT-4 are estimated at between 78 and over 100 million US dollars, while newer models like Gemini Ultra are expected to incur training costs of up to 191 million US dollars.

This cost discrepancy is revealing. While the human labor required to create the training data remains virtually unpaid, investments flow into computing power, hardware, and highly skilled technical personnel. A study by researchers at the University of Toronto and Chapel Hill calculated what it would cost if the human labor contained in training data were fairly compensated. Even under very conservative assumptions, the hypothetical costs of data creation exceed the actual training costs tenfold to a thousandfold. For GPT-4, the value of the data used would thus be over US$30 billion; for newer models, it could be significantly higher. These figures illustrate the extent of the value shift: all of humanity's creative and informational labor is being turned into free input for business models whose profits remain concentrated in the hands of a few corporations.

The argument of AI companies that their models merely learn from data and do not create copies obscures this economic reality. Even assuming, technically, that a trained model does not store exact copies, the fact remains that these models would not function without the creative contributions of millions of authors. The parameters of a neural network are the distilled result of processing these works. They represent the extracted value from human creativity. In this respect, it is a form of appropriation that, while technologically mediated, economically resembles classic expropriation.

Memorization as a technical and economic problem

The technical debate surrounding the concept of memorization is central to its legal and economic evaluation. Research has demonstrated that large language models are indeed capable of reproducing training data verbatim, particularly when certain prompting techniques are employed. A study by Google DeepMind and other institutions showed that ChatGPT, using a simple trick where the model was prompted to repeat a word, suddenly output several megabytes of training data, even though the model was designed to prevent this. The researchers extracted several megabytes of memorized content, including personal information, copyrighted texts, and other sensitive data, at a cost of approximately two hundred US dollars.

These findings contradict OpenAI's claim that the model does not store data. Memorization occurs particularly when certain text sequences appear very frequently in the training dataset. Popular song lyrics that have been repeated on countless websites are practically predestined for this effect. The model learns not only abstract language patterns but also concrete sequences that it can retrieve when given corresponding input. The distinction between learned patterns and stored data thus becomes blurred. From a legal perspective, the crucial point is that copyrighted content is being output, regardless of how this output is technically generated.

From an economic perspective, memorization means that the value created by the original texts is directly transferred into the model. ChatGPT can provide users with song lyrics without requiring them to visit the GEMA website or other licensed sources. This represents a direct substitution that deprives rights holders of potential revenue. While search engines like Google redirect users to the original sources, thereby generating traffic that can be monetized, ChatGPT terminates this value chain. The user receives the information directly from the model, leaving the copyright holder empty-handed. This form of disintermediation is a core feature of many platform business models, but here it reaches a new level because it directly impacts the creative process itself.

Benefit from Xpert.Digital's extensive, fivefold expertise in a comprehensive service package | R&D, XR, PR & Digital Visibility Optimization - Image: Xpert.Digital

Xpert.Digital has in-depth knowledge of various industries. This allows us to develop tailor-made strategies that are tailored precisely to the requirements and challenges of your specific market segment. By continually analyzing market trends and following industry developments, we can act with foresight and offer innovative solutions. Through the combination of experience and knowledge, we generate added value and give our customers a decisive competitive advantage.

More about it here:

• Use the 5x expertise of Xpert.Digital in one package - starting at just €500/month

Asymmetries of bargaining power

The dispute between GEMA and OpenAI is embedded in a fundamental power imbalance between the technology sector and the creative industries. OpenAI possesses virtually unlimited financial resources: In 2025 alone, the company plans expenditures of approximately eight billion US dollars, and by 2030, cumulative investments in infrastructure, training, and personnel are expected to reach nearly 100 billion US dollars. These funds come from investors such as Microsoft, SoftBank, and other capital providers who anticipate a fifty-fold increase in revenue by 2030. At the court hearing in Munich, seven lawyers and two legal counsel represented OpenAI—a legal force that far exceeds the resources of even large collecting societies.

On the other side are creative professionals whose incomes are already under considerable pressure due to the streaming economy. Studies on music streaming in Germany show that 68 percent of artists earn less than one euro a year from their streamed works. Revenues are extremely concentrated: 75 percent of earnings go to just 0.1 percent of artists. The business model of streaming platforms, in which artists are not paid for actual streams but rather for their share of the total number of streams, systematically disadvantages small and medium-sized artists. Into this already precarious situation, generative AI is now encroaching, threatening to occupy even those market niches previously occupied by humans.

The bargaining power of the creative industries is structurally limited. Unlike in industrial production, where unions and collective bargaining agreements provide a degree of balance, comparable mechanisms are lacking in the cultural sector. Collecting societies like GEMA do play an important role, but they rely on enforcing existing rights. However, when the legal situation is unclear and courts only provide clarification after years, a de facto situation arises in which technological development creates facts that are virtually impossible to address legally. It could be years before the Munich ruling becomes legally binding. During this time, ChatGPT will continue to be used by hundreds of millions of people, OpenAI will expand its market position, and the acceptance of AI-generated content will increase.

This asymmetry is also evident in the political arena. Large technology companies wield considerable influence over political decision-making processes through lobbying, the threat of relocation, and the narrative that regulation stifles innovation. While the European Union's AI Regulation, which partially entered into force in August 2025, obliges providers of general-purpose AI models to be more transparent about the training data they use, the concrete implementation of these requirements remains the subject of intense negotiations, in which industry is attempting to secure the broadest possible exemptions and transitional periods.

The GEMA licensing model as a counter-model

In response to systematic non-payment, GEMA became the first collecting society worldwide to introduce a licensing model for generative AI in September 2024. This two-pillar model aims to capture value at both points where it arises: during the training of the models and during the use of the generated content. The first pillar is aimed at the providers of the AI ​​systems and provides for a 30 percent share of all net revenues generated by the model. This includes subscription fees, license fees, and other income. Additionally, a minimum fee will apply, based on the volume of generated content, to include models that generate little direct revenue but are nevertheless widely used.

The second pillar concerns the subsequent use of AI-generated music content. If, for example, a song created with an AI tool is used on streaming platforms, in advertising, or as background music, royalties should also flow to the creators of the original works used for training. This model recognizes that the value chain does not end with training, but that the generated content itself is commercially exploited and competes with human-created music.

GEMA's justification for the level of the requested contribution is noteworthy. They argue that the use of original works for generative AI purposes represents the most intensive form of use imaginable. Unlike a single reproduction or performance, where the work retains its identity, AI transforms it into raw material for generating new content that can replace or displace the original. The creative work of the authors forms the indispensable basis for the entire economic success of AI providers. Against this backdrop, a 30 percent contribution does not appear excessive, but rather an attempt to secure a fair share of the added value.

Critics of the model, primarily from the technology sector, warn of a stifling innovation. They argue that licensing costs could hinder the development of new AI applications and set Europe back in international competition. This argument, however, overlooks the fact that innovation is not synonymous with the free appropriation of others' work. Even in the pharmaceutical industry, where research and development are extremely expensive, the argument is not that one should therefore be free to use patented substances. The real question is how the costs and benefits of technological progress are distributed and whether an economic system is acceptable in which a few corporations reap astronomical profits while the creative individuals on whose work everything depends are systematically left empty-handed.

The international dimension and comparable conflicts

The Munich case is not an isolated incident, but part of a global dispute. In the US, several authors' associations, publishers, and media companies have filed lawsuits against OpenAI and other AI providers. The New York Times sued OpenAI and Microsoft in December 2023, accusing the companies of using millions of articles for training purposes without permission. Other cases concern the use of books, scientific publications, and program code. In February 2025, a US federal court ruled for the first time that using copyrighted data to train an AI can constitute copyright infringement, even if the developer was unaware of the specific infringement.

In Europe, the Budapest District Court has referred questions to the European Court of Justice (ECJ) regarding Google Gemini's use of copyrighted content. The case concerns an article about a planned dolphin aquarium, which the chatbot reproduced almost verbatim. The Hungarian lawsuit addresses both copyright and the related rights of press publishers. The ECJ will have to clarify whether the reproduction of content by a chatbot constitutes reproduction and making available to the public within the meaning of EU law, and what role the fact that the models are based on probabilistic predictions plays. This referral is the first of its kind on the topic of generative AI and will set a precedent for the entire European Union.

The international dimension demonstrates that this is a systemic conflict that cannot be resolved through isolated national rulings. AI models are trained globally, the training data originates from all over the world, and their use is cross-border. A fragmented legal framework in which each country sets its own standards would lead to considerable uncertainty. At the same time, there is a risk that large platforms will engage in regulatory arbitrage by shifting their activities to jurisdictions where copyright enforcement is weakest. GEMA deliberately chose to file its lawsuit in Munich because it has a chamber specializing in copyright law, increasing the likelihood of an expert decision.

Future scenarios and systemic decisions

The Munich ruling will not be the final word in this dispute. Both sides have already announced that they expect the case to be referred to the European Court of Justice if it goes to appeal. Only a fundamental decision at the European level can clarify the numerous open legal questions arising from the use of copyrighted works by AI. Central to this are questions such as: Does the training of AI models fall under the text and data mining exception, or is it a use requiring a license? Is the output of content by a chatbot an independent copyright infringement? How should the memorization of data be assessed from a technical and legal perspective? And what requirements must be met for an effective reservation of rights?

The answers to these questions will fundamentally influence the business models of the AI ​​industry. Should the courts conclude that licenses are required, companies would either have to raise substantial sums to acquire usage rights or train their models using licensed or synthetic data. Both options would significantly increase costs and could alter the market structure. Smaller providers, lacking the financial resources of large corporations, could be squeezed out of the market, leading to even greater concentration. On the other hand, legally secure licensing would also open up new business opportunities, for example, for collecting societies, database providers, and content brokers who act as intermediaries between rights holders and AI developers.

An alternative scenario involves policymakers finding regulatory solutions that balance promoting innovation with protecting copyright. The EU AI Regulation already imposes transparency obligations on AI providers, who must disclose which data they used for training. A next step could be a legally mandated remuneration system, where AI providers pay a flat fee that is then distributed to rights holders according to a predetermined formula. This model would reduce bureaucracy and enable widespread use of training data without the need to negotiate licenses on a case-by-case basis. However, the amount of such a fee and the distribution mechanisms would be highly controversial politically.

A third scenario is the emergence of new collective bargaining structures. Similar to unions for workers, associations of creators could form, giving them greater leverage vis-à-vis the platforms. Some initiatives in this direction already exist, such as the Coalition for Content Provenance and Authenticity, which advocates for the labeling of content, or projects to develop opt-out standards that make it easier for rights holders to exclude their works from training. However, the effectiveness of such initiatives depends on support from legislation and jurisprudence.

The Reassessment of Creative Capitalism

The Munich Regional Court's ruling is more than just a legal decision about nine song lyrics. It marks the beginning of a necessary societal debate about who deserves the fruits of digital transformation and according to which principles value creation should be organized in the age of artificial intelligence. In recent years, technology companies have created a reality in which the free appropriation of creative work has become the foundation of gigantic business models. This practice could be maintained as long as the legal situation remained unclear and the affected creative professionals lacked effective means of redress.

The Munich ruling changes this situation. It establishes that the existing legal framework, created to protect human creativity, remains valid even in the age of AI. The argument of technology companies that their models only learn and do not create copies is seen as a smokescreen obscuring the true economic realities. The question is not whether AI memorizes in a technical sense, but whether the use of others' works for training and the subsequent output of these works results in a shift in value in favor of the platforms and at the expense of the copyright holders. The answer is obvious.

The coming years will show whether this ruling marks the beginning of a realignment of power dynamics or whether it remains a symbolic victory unable to halt actual developments. The history of digitalization is replete with examples where courts established rights that were then practically unenforced because technological and economic dynamics outweighed the law. Crucially, policymakers will have the courage to create clear frameworks that ensure fair participation for creative professionals without stifling innovation. This is no easy task, but it is essential if we want to prevent cultural production from being subjected solely to the economic imperatives of a few corporations.

In the long historical perspective, the Munich ruling is part of a series of other debates surrounding the appropriation of the commons. Like the enclosure of the commons during the transition to a market economy or the privatization of public goods under neoliberalism, the central question here is what belongs to the public and what may be appropriated by private enterprise. Humanity's creativity, embodied in millions of works, is a collective good. The question of whether a few corporations should be allowed to transfer this good into exclusive business models free of charge touches upon the core of our economic order. The Munich ruling is a step towards an answer that takes the rights of creators seriously. Whether this step will be sufficient remains to be seen.

☑️ Our business language is English or German

☑️ NEW: Correspondence in your national language!

Konrad Wolfenstein

I would be happy to serve you and my team as a personal advisor.

You can contact me by filling out the contact form or simply call me on +49 89 89 674 804 (Munich) . My email address is: wolfenstein ∂ xpert.digital

I'm looking forward to our joint project.

☑️ SME support in strategy, consulting, planning and implementation

☑️ Creation or realignment of the digital strategy and digitalization

☑️ Expansion and optimization of international sales processes

☑️ Global & Digital B2B trading platforms

☑️ Pioneer Business Development / Marketing / PR / Trade Fairs

Our global industry and business expertise in business development, sales and marketing - Image: Xpert.Digital

Industry focus: B2B, digitalization (from AI to XR), mechanical engineering, logistics, renewable energies and industry

More about it here:

• Xpert Business Hub

A topic hub with insights and expertise:

• Knowledge platform on the global and regional economy, innovation and industry-specific trends
• Collection of analyses, impulses and background information from our focus areas
• A place for expertise and information on current developments in business and technology
• Topic hub for companies that want to learn about markets, digitalization and industry innovations