On 15 September, the Danish Ministry of Culture published a report from the Expert Group on Copyright and Artificial Intelligence (AI). This article describes parts of the report and comments on certain ambiguities in the report.
The report from the Expert Group on Copyright and Artificial Intelligence (AI) was published by the Ministry of Culture on 15 September 2025. The report contains several recommendations aimed at addressing the challenges posed by artificial intelligence in relation to copyright.
The report thus includes recommendations for improved transparency and control over training data, strengthened framework conditions for collective licensing and the introduction of technical measures to prevent illegal use of copyright-protected content. In addition, an investigation into measures to promote the use of human-created content is proposed together with guidance and awareness initiatives to ensure clarity and legal certainty in the use of AI systems.
The Expert Group consisted of representatives of the Joint Council on Copyright (Samrådet for Ophavsret), The Danish Rights Alliance, the Danish Chamber of Commerce, Danish Industry, the Association of Danish Media (Danske Medier), the Royal Danish Library as well as technical and legal experts.
The report was long awaited by numerous people in the creative sector, etc., and since, according to the terms of reference, it should have been submitted to the Danish Minister of Culture already in the winter of 2024/2025.
The report marks an important step in the Danish handling of AI in relation to copyright and will undoubtedly have an impact. However, the report contains some ambiguities and does not seem to be completely updated, which this article intends to highlight.
Code relating to the AI Regulation
Paragraph 1.2 of the report regarding the AI Regulation states (p. 24) that "the Code was planned to be finalised on 2 May 2025. However, the publication of a final Code of Practice has been postponed".
It should be noted that the Code was published on 10 July 2025 and approved by the European Commission and the AI Committee on 1 August, just before the AI Regulation's rules on AI for general use came into force on 2 August 2025. This was earlier than the Expert Group's report from September.
The Expert Group has not included the Code's rules in its assessment, which would have been relevant even though the Code constitutes soft law in the sense that the rules are formally non-compulsory. This means that the report does not appear to be completely up-to-date.
Information on training data
The Expert Group's recommendation 1 deals with "effective transparency in training data" and that effective transparency is ensured at EU level when using copyright-protected content to train and develop commercial AI systems.
It is noted that Article 53(1)(d) of the AI Regulation lays down a requirement that providers must prepare and publish a sufficiently detailed summary of the content used to train the model. This training summary must, according to the overall AI set of rules, be made according to a template provided by the so-called AI Office, which is also mentioned in Recital 107 of the AI Regulation.
Users' illegal uploads
The Expert Group addresses the situation where users copy copyright-protected material for instructions or upload such material to the systems. Similarly, recommendation 6 made by the Expert Group concerns "requirements of technical measures to prevent illegal uploading and copying of copyright-protected content on AI services".
In this context, it is worth noting the AI Code's requirement for providers to prohibit copyright-infringing use in their terms of use.
The Expert Group also mentions (p. 49) that many large tech companies have clauses in their terms and conditions that "imply that the user is liable for potential copyright infringement if the user uploads copyright-protected material without authorisation from the rightsholder". This raises the question of whether technical measures are necessary to prevent illegal uploading and copying of copyright-protected content on AI services.
In relation to training of the AI system, the Expert Group (p. 50) mentions that users may contribute to the ongoing training of the system without being conscious or aware of it.
It is noted that, according to the AI Code, providers are required to implement technical security measures to ensure that the model does not reproduce the training content in its output to the effect that this constitutes copyright infringement.
Do Articles 3-4 of the DSM Directive relate to AI training?
The report (p. 22) states that "Artificial intelligence is not mentioned in the DSM Directive in relation to text and data mining. The question of whether and to what extent these exemptions apply to training and other uses of artificial intelligence is therefore unresolved." However, the Expert Group states that the AI Regulation, which came after the DSM Directive, "may indicate that the EU legislator perceives the rules on text and data mining under Article 4 of the DSM Directive as a possible (future) legal basis for model training".
Later on in the report (p. 38), the Expert Group also states that "Whether artificial intelligence training may be equated with text and data mining is controversial". One might ask whether the Expert Group believes that it is questionable whether the text and data mining rules in ss 11b-c of the Danish Copyright Act (ophavsretsloven) may be applied to AI model training.
The confusion is underlined by the fact that, even before this (earlier on in the report on p. 38), the Expert Group applies the text and data mining rules to AI: "The provision in s. 11b(2) makes it possible, among other things, for developers of artificial intelligence to use copyright-protected material and create copies of works and other creations for the purpose of text and data mining without requiring authorisation from the rightsholders".
We note that it was indeed unclear/discussed after the DSM Directive in 2019 and became particularly relevant with the launch of ChatGPT in November 2022 whether Articles 3 and 4 of the DSM Directive related to AI training. With the adoption of the AI Regulation in March 2024, there seems to be no question in wider circles that text and data mining in the DSM Directive is the same as AI training, see Article 53(1)(c) of the AI Regulation, which clearly expresses this by directly in its wording referring to "Article 4(3) of Directive (EU) 2019/790", which is the AI Regulation.
Training prior to 2 August 2025
The report (p. 39) states that the AI Regulation "does not, however, affect training that took place before the Regulation came into force".
However, the decisive factor is not the time of training, but when the model is placed on the market, i.e. training taking place in 2024, for example, is covered by the AI Regulation if the model is placed on the market after 2 August 2025. Furthermore, transitional rules mean that even models trained and placed on the market before 2 August 2025 must comply with the AI Regulation, only with a two-year longer deadline.
Article 111(3) of the AI Regulation thus states that "Providers of general-purpose AI models placed on the market before 2 August 2025 must take the steps necessary to comply with the obligations set out in this Regulation by 2 August 2027."
Digital imitations of personal characteristics
Recommendation 5 made by the Expert Group concerns "protection against digital imitations of individuals' personal characteristics".
It should be noted that a bill in that respect has already been submitted for public consultation by the Ministry of Culture with a consultation deadline on 21 August 2025, in which connection the Ministry stated that the bill is expected to be enacted on 31 March 2026. This bill is not mentioned in the report by the Expert Group.
Significance for enterprises
Enterprises should be aware that, as mentioned above, the Expert Group's report contains some ambiguities and is not fully updated on current rules/developments, as it does not include the AI Code, which is relevant for recommendation 1 (effective transparency in training data) and recommendation 2 (effective opt-out mechanisms or revised rules for text and data mining).
Recommendation 5 (protection against digital imitations of individuals' personal characteristics) and Recommendation 6 (requirement of technical measures to prevent illegal uploading and copying of copyright-protected content on AI services) are discussed above.
The report's other recommendations are immediately relevant. These are strengthening the framework for collective licensing (recommendation 3), pilot scheme for mandatory arbitration in press publication rights disputes (recommendation 4), conditional public prosecution in copyright and AI cases that are technically and territorially complex (recommendation 7), exploring possible measures to promote the use of human-created content (recommendation 8), guidance and awareness initiatives on copyright and AI (recommendation 9), and clarification in copyright law that the provision of AI systems constitutes communication to the public (recommendation 10).
To read the report drawn up by the Expert Group, click here.
Read our article on the AI Code of Practice here.
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.
