- within Intellectual Property topic(s)
- with readers working within the Banking & Credit, Technology and Law Firm industries
To nye europæiske domme om generativ AI og ophavsret viser, at både træningsdata og AI-genereret output kan få afgørende betydning i kommende sager. For danske virksomheder er det et område, der bør håndteres som en konkret juridisk risiko – ikke kun som et teknologisk spørgsmål.
In November 2025, judgments were delivered in two high-profile European cases concerning generative AI models and copyright: Getty Images vs Stability Al before the High Court of England and Wales, and Gema vs OpenAI before the Munich Regional Court.
The two cases are not Danish, but they are relevant to the Danish corporate sector in that they touch upon some of the issues which are already in focus when advising on AI: Which data was used to train the Al model? Have the reservations made by rightsholders as to text and data mining been considered? Does a risk exist that the model’s output may reproduce or closely resemble copyrighted material? And how is this risk to be addressed in contracts, internal policies and the commercial use of AI?
The two cases had very different outcomes. In the Getty case, the High Court did not find that Stable Diffusion itself contained an infringing copy of Getty’s images. In the Gema case, however, the Regional Court found that OpenAI had infringed copyright, partly due to the fact that song lyrics were reproducibly contained in the model and could be reproduced in the output.
For businesses, the key point is therefore not only that the legal position remains unclear. The point is also that this uncertainty should already be translated into specific decisions as to the use, procurement, development and management of AI solutions.
The Getty case: Training is not copying
The Getty case concerned Stability AI’s image-generating AI model, Stable Diffusion. Getty Images argued that the model had been trained on millions of Getty’s copyrighted images collected from the Internet without consent.
However, the High Court found that Stable Diffusion did not store or reproduce Getty’s works in a way that constituted an infringing reproduction under British copyright law. The High Court stressed that the model weights did not themselves store Getty’s protected works, but instead represented the patterns and features learned by the model during training.
The Gema case: Output and memorisation may give rise to liability
The Gema case dealt with music rights and the lyrics of nine popular German songs. The German rights management organisation Gema brought proceedings against OpenAI, alleging that ChatGPT had been trained without permission based on song lyrics from Gema members and could reproduce those lyrics almost word for word when users asked for specific works.
The Regional Court found that copyright had been infringed in several ways. The Court emphasised that the disputed song lyrics were incorporated into OpenAl’s language models in a reproducible manner, and that memorisation could be established by comparing the song lyrics in training data with the model’s output.
The Court further held that the chatbot’s output constituted unlawful reproduction as well as making copyrighted song lyrics available to the public. OpenAI was therefore ordered, as the model provider, to pay compensation, although the amount was not disclosed.
What are the implications of the cases in a Danish context?
The Getty and Gema cases have not been decided under Danish law, but they highlight issues that are also likely to be key to Danish cases on generative AI and copyright.
Many of the most widely used generative AI models in Denmark originate in the US and are typically trained outside Denmark. However, this does not rule out that material owned by Danish rightsholders may have been included in the training. Consequently, issues relating to jurisdiction, governing law and the use of material under copyright law may prove to be of great significance in future cases.
This is already the case in Denmark. In November 2025, Koda brought legal proceedings against the American AI music services provider Suno and, in February 2026, DPCMO brought legal proceedings against LinkedIn and OpenAI, respectively. The Koda case is similar to the Gema case in several respects as it focuses on specific works and allegations that the output is identical or nearly identical to copyrighted material that has allegedly been used for training without consent.
These cases therefore highlight the need for Danish businesses to clarify their own roles: Is the business in question a rightsholder, an AI provider, a supplier, a customer or a user of AI output? The answer determines which risks are most relevant to focus on, and which contractual and organisational measures should be given top priority.
Text and data mining, opt-outs and contracts
A key issue in future cases will be how the rules on text and data mining and the rightsholders’ opt-outs are to be interpreted in an AI context.
Depending on the circumstances, the rules on text and data mining may allow for certain copying to be carried out as part of data analysis, but rightsholders may reserve the right to object to such use. In the Gema case, the Court found that the text and data mining exemptions did not apply because the model not only extracted information but also reproduced works in a way that interfered with the rightsholders’ exploitation interests.
The AI Regulation also contains a recital stating that providers placing general-purpose AI models on the EU market should comply with EU copyright rules, regardless of where the copyright-relevant acts involved in training took place. The exact scope remains a matter of debate, but the provision suggests that businesses cannot disregard EU rules simply because the training took place outside the EU.
For clients, this makes the contractual basis a key consideration. When procuring, developing or using AI solutions, contracts should address, for example:
-
what information the supplier may provide on training data and rights management
-
whether rightsholders’ opt-outs are observed
-
who assumes the risk of third-party claims
-
how AI-generated output may be used
-
what warranties, limitations of liability and indemnities apply
-
what documentation and audit rights are available to the customer
The practical benefit is that the business may clarify responsibilities, reduce uncertainty and establish an improved basis for dealing with claims in the event that a copyright dispute arises at a later date.
Current actions to be taken by the corporate sector
The corporate sector should not wait for definitive legal guidance before taking a position on copyright-related AI risks. National courts and the CJEU are expected to provide further guidance in the coming years on issues such as model training, outputs, text and data mining, opt-outs and the liability of AI providers. However, the risk is already a practical concern.
AI model developers and providers should have procedures addressing training data, opt-outs, documentation, risk assessment and the prevention of memorisation or reproduction of protected material.
Procurers of AI solutions should review supplier agreements to ensure clear regulation of rights, liability, documentation and the handling of third-party claims.
Users of generative AI tools should establish internal policies governing input, output, approval and external use to ensure that employees do not upload unauthorised material or inadvertently use AI-generated material in a way that may infringe the rights of third parties.
At the same time, rightsholders should consider how rights reservations are worded and documented, and how potential infringements may be substantiated through specific examples of output.
Our comments
The Getty and Gema cases highlight the continuing significant legal uncertainty surrounding AI and copyright. This applies both to the question of when training AI models involves copying of copyrighted material, and when AI-generated output may constitute an infringement in its own right.
The cases further indicate that text and data mining rules and rightsholders’ opt-outs are likely to become key issues in the coming years. These issues remain open and are likely to become subject to further judicial scrutiny at both national and EU level.
For businesses, this means that these issues should already be approached as specific legal and commercial risks as of now. This is relevant to rightsholders, technology companies and businesses procuring or using generative AI solutions in their commercial activities.
The value of the advice is in translating legal uncertainty into practical, actionable measures. Businesses should, among other things, maintain an overview of AI use, clear contractual terms on liability and rights, internal guidelines for the use of AI tools, procedures for verifying outputs and documentation showing how rights and opt-outs are managed.
In the absence of clear rules, businesses are best protected by ensuring transparency, clear responsibility allocation and processes designed to reduce the risk of copyright disputes.
We have extensive experience advising on copyright, media law and new forms of digital exploitation of protected content. We advise on the legal framework governing the development and use of AI solutions and assist on strategic assessments, contractual regulation and the handling of specific copyright issues.
The full article on AI and copyright: The Getty and Gema cases in a Danish context has been published in Nordiskt Immateriellt Rättsskydd (NIR) No. 2/2026.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
[View Source]