Artificial Intelligence (AI) is driving innovation across industries, but it also presents legal challenges, including in the area of copyright.
In Switzerland, recent discussions at the federal level, such as the Federal Council’s AI framework and a motion adopted by the Council of States, indicate a growing awareness of the need to address copyright considerations related to AI. While these developments do not yet amount to binding legal changes, they suggest that adjustments to the Swiss Copyright Act (CopA) may be considered in the future.
This article provides a practical overview of the current legal status, key areas of uncertainty, and potential changes to Swiss copyright law.
Potential Changes to the Copyright Act
Framework for the regulation of artificial intelligence of the Federal Council
In November 2023, the Federal Council commissioned the Federal Department of the Environment, Transport, Energy and Communications (DETEC) and the Federal Department of Foreign Affairs (FDFA) to prepare an overview of possible regulatory approaches to AI in Switzerland, which they published in early 2025. The published overview includes an underlying legal baseline analysis, which provides a comprehensive legal analysis of regulatory approaches to AI in Switzerland and also specifically addresses the intersection of copyright law and AI.
The analysis acknowledges the present uncertainties surrounding the use of AI systems and associated copyright issues. Key questions are whether AI-generated results are protected by copyright and if training AI is relevant under copyright law. The analysis concludes that if training AI is deemed copyright-relevant, regulatory measures must be considered to ensure AI development. Conversely, if it is not, the interests of rights holders must be addressed by the regulator. Consequently, this implies that there will inevitably be some form of adjustment to the CopA. However, the Federal Council has not commented on any intentions to amend the CopA.
Pending motion on the protection of IP from AI misuse
In December 2024, a member of the Swiss Council of States submitted a motion titled “Better Protection of Intellectual Property from AI Misuse.” The motion calls on the Federal Council to ensure that works protected under Swiss copyright law are comprehensively safeguarded when used by providers of AI systems. The motion demands that rights holders must give explicit permission before their works can be used to train generative AI systems. It further demands that providers of publicly available AI systems should not be allowed to invoke existing exceptions in the CopA and that if AI systems are made available in Switzerland, Swiss law shall apply, and Swiss courts shall have jurisdiction.
The Federal Council supported the motion, and the Council of States adopted it on March 20, 2025. The motion now proceeds to the National Council for consideration. While the motion has passed an important milestone in the Council of States, its future remains uncertain. The National Council may still reject or amend it, and even if adopted, the process of translating the motion into binding law is time-consuming and includes a public consultation process. Given the typical pace of Swiss legislative procedures, it is unlikely that a new law will be in force before 2027. The drafting, consultation, and parliamentary approval phases can each take several months to over a year, especially for legislation involving emerging technologies and international implications.
Current legal landscape in Switzerland
AI Training
One of the primary questions currently discussed among scholars and practitioners is whether the use of copyrighted material to train AI models involves acts of copying which are capable of constituting copyright infringement. Experts remain divided on this issue.
It is important to subcategorize the term “training” into (i) the collection of training data and (ii) the use of such data for the training process. This distinction is relevant as these actions are often conducted by different economic actors. Regarding the first step, the creation of training datasets, there seems to be consensus that this action – absent consent of the author or the application of a limitation – constitutes copyright infringement as it involves copying, which falls under the reproduction rights of copyright holders. This includes downloading and storing large amounts of data, often using automated tools like web crawlers and scrapers.
Regarding the second step, the actual process of training, different views are taken. Some authors argue that the actual training process may not involve a potentially infringing act, depending on whether and how the copies are made. Others take the view that the concept of reproduction has to be interpreted broadly, including copies that merely enable potential (human) consumption, regardless of whether actual consumption of the work occurs. Based on this latter view, the actual training process, if it requires reproductions of the underlying training data, could constitute a copyright infringement. Lastly, at the other end of the spectrum, there are authors who conclude that the training process is not capable of infringing copyrights because using copyrighted content by the AI model for training does not lead to human perception of a work and thus falls outside the scope of the perception of a work under copyright law.
Assessing whether there has been a potentially infringing act of copying will always be a fact-specific inquiry dependent on exactly how the training dataset was created and what is involved in the training process from a technical perspective. To date there seem to be no relevant cases pending in Switzerland that could give any answers to this question. If Swiss courts determine that training of an AI system involved reproductions that are relevant under copyright law, then permissions from the rights holders would need to be obtained, unless an exception applies.
Exceptions and Limitations
Swiss copyright law provides several exceptions that might apply to AI training, but these are often limited in scope:
Internal business use: The internal business use exception allows companies or organizations to use copyrighted works for internal use (Art. 19 para. 1 let. c CopA). However, this exception only applies in cases where the copyrighted works are reproduced solely for internal information or documentation. Many argue that this is not the case for the majority of AI training, the purpose of which is external use. If the exemption applies, the full or nearly full copying of commercially available works, visual art, music scores, and software is not allowed under the exception. Furthermore, businesses have to pay a compensation rate determined by the common tariff system for reproductions they make internally under the exemption. Temporary copies: The exception for temporary copies permits reproductions, that are transient or incidental, integral to a technical process, solely for transmission in a network between third parties by an intermediary or for lawful use, and that have no independent economic significance (Art. 24a CopA).
While some reproductions during AI training might be considered temporary and part of the technical process within the model, lawful use (i.e. consent or the application of another exception) may not be given. Some authors further believe that the copies made in training have independent significance.
Scientific Research: The scientific research exception allows the use of copyrighted works for research purposes (Art. 24d CopA). This exception is broader in Switzerland than in some other jurisdictions such as e.g. the EU data mining exception for the purposes of scientific research, as it covers both public and private research. However, it applies only if the primary purpose of the AI training is scientific research. While a parallel commercial purpose is allowed, for the exception to apply, the primary purpose must remain scientific research. This might often not be the case for commercial AI applications. In addition to the primary scientific purpose, for the scientific research exemption to apply, the reproduction must be integral to a technical process and the access to the reproduced works must be legal. The reproduction for a technical process is considered integral in the context of AI training. Whether the access to the reproduced works is legal, particularly with regard to publicly available works on the internet, is subject to discussion among scholars. For clarity, Switzerland does not have a broad exception or limitation for text and data mining (such as Art. 4 of Directive 2019/790/EU).
Against this background, it cannot be generally stated that an exception can be applied to AI training. Whether an exception can be applied to the training of a specific AI model would need to be examined on a case-by-case basis.
Ownership of AI-Generated Output / AI-Assisted Output
Swiss copyright law currently only protects works created by humans (Schöpferprinzip). Therefore, if an AI system generates content, this output is not protected by copyright unless a human has made a significant creative contribution. This stance aligns with the broader European perspective, where human authorship is a prerequisite for copyright protection.
Most experts in the field want to recognize an intellectual creation when a natural person determines the AI-supported creation process and significantly defines the quality characteristics required for the protectability of the intellectual creation, i.e., the individual character of a work in copyright law. Simply put: when AI is used merely as a tool, and the human user exercises creative control, the resulting work can be copyrighted. Conversely, if the AI independently generates the content, it lacks copyright protection. Where the line must be drawn from a practical perspective remains to be seen and will have to be assessed on a case-by-case basis taking into account the creation process of the work.
There is consensus, for example, that a purely machine translation of a text is not protected by copyright, as neither the developers of the translation tool nor the user have performed an intellectual creative act. However, a translation done by a natural person would be considered an intellectual creation and an independently protected derivative work, provided it has individual character.
Copyright Infringement of AI-Generated / AI-Assisted Output
Regarding the output of an AI system, there is not only the question of copyright protection but also whether the output itself could infringe on copyrights. This would be the case if the AI result reproduces a work unchanged or in a way that the original work is still recognizable. Whether this is the case must be determined positively considering the facts on each individual case.
Territoriality
The principle of territoriality in copyright law means that copyright protection is generally confined to the borders of the country where the copyright is granted. In Switzerland, the territoriality principle means that Swiss copyright law applies only to activities within Swiss borders. This principle poses challenges in potential prosecution of copyright infringements in the context of AI, as AI systems often operate across multiple jurisdictions.
Conclusion
In conclusion, as is the case for many jurisdictions around the world, the intersection of AI and copyright law in Switzerland presents an evolving landscape. The primary unresolved issue is whether the second step in the AI training process, namely the actual learning of the AI model based on training data sets, constitutes a copyright-relevant act. This determination depends on a nuanced understanding of the technical processes involved, which can vary significantly between different AI models. The current legal framework offers some exceptions and limitations, but their applicability to AI training remains uncertain and often requires case-by-case analysis.
The Federal Council’s publication of an AI regulatory framework and the Council of States’ adoption of a motion addressing copyright concerns reflect a growing awareness of the legal implications of AI. While these developments do not yet amount to concrete legal changes, they suggest that copyright issues related to AI may receive more focused attention in the coming years. Should the motion also pass in the National Council, it could eventually lead to targeted amendments to the CopA, potentially clarifying the need for rights holders’ consent in AI training and addressing jurisdictional questions.
