The Ministry of Culture publishes the expert group’s recommendations on copyright and artificial intelligence

Effective transparency is a prerequisite for controlling the use of content in the development of AI services.

The Rights Alliance has documented here that the new rules adopted by the European Commission are not sufficient for rights holders to verify whether their content is being used to train AI models.

The Rights Alliance has also documented here that the Commission’s recent implementation in the so-called Code of Practice and the transparency template for AI providers is not sufficient for rights holders to verify whether their content is being used to train AI models.

The expert group recommends that the Danish government work within the EU to ensure effective transparency regarding the use of copyright-protected content for training and developing artificial intelligence. Effective transparency can be achieved either by requiring AI providers to make certain information available to rights holders, or by granting rights holders direct access to search the datasets used for training.

On page 57 of the report, the expert group describes the specific information AI providers must make available to achieve effective transparency.

If AI providers do not meet the requirements for effective transparency, a rule of reversed burden of proof should apply, with the presumption that rights holders’ content has been used. It would then be the responsibility of the AI provider to prove otherwise.

The Rights Alliance believes that the current exception allowing the use of content for text and data mining turns the copyright system upside down, as rights holders are forced to ‘opt out’ instead of giving consent to the use of their content. Furthermore, the Rights Alliance notes that the current opt-out system does not work in practice, as rights holders have no way to issue reservations that are effective or respected by AI service developers.

The expert group recommends, as a first step, implementing feasible solutions that will improve the effectiveness of the current opt-out mechanisms.

If these challenges are not resolved by the end of 2026, Denmark should push for Article 4 of the DSM Directive to be amended as quickly as possible so that it is based on an opt-in system. This would mean that the use of works for text and data mining would require the explicit consent of the rights holder. The assessment of this should take place in connection with the planned evaluation of the DSM Directive in 2026.

The expert group believes that rights holders are entitled to receive fair remuneration when their content is used for the commercial development and training of AI. A stronger foundation must be established to make this possible, through collective licensing solutions.

The expert group recommends that both Denmark and the EU take initiatives to improve the framework conditions for achieving collective licensing solutions that ensure more transparent and fair agreements with AI providers.

At the EU level, legal regulation should be introduced to ensure that collective management organizations can issue licensing solutions to AI providers (potentially through larger global schemes).

In Denmark, the possibility of license agreements can be further strengthened through an extended collective licensing provision covering the use of copyright-protected material in AI systems, which would also mean that disputes could be resolved by the Copyright License Tribunal. In addition, the expert group encourages the facilitation of collaboration between stakeholders involved in developing Danish language models.

It is a major challenge to encourage tech companies to enter into agreements with rights holders regarding the use of content. Today, there are mediation models based on voluntariness, but these are not effective in practice, partly because there are no consequences if parties fail to attend mediation.

The expert group recommends examining whether a mandatory arbitration model could be an effective solution to rights disputes and promote agreements on the use of protected content.

Publishers of press publications are particularly affected by the difficulties in getting tech companies to enter into agreements on the distribution of press publications on information society services. Therefore, the expert group recommends that a pilot scheme first be tested in this area.

Rights holders must be better protected against the distribution of digital imitations of their voice, appearance, and other personal characteristics.

The expert group therefore recommends introducing new rules in the Copyright Act on the protection of personal characteristics, codifying parts of personality rights that are currently difficult to enforce against online platforms. The expert group stresses that swift action is needed, and that such rules should therefore first be implemented at the national level.

The Rights Alliance highlighted the problems of lacking enforcement against voice clones and similar technologies in spring 2024 here. Since then, the Minister of Culture and the government have proposed introducing rules in the Copyright Act that establish general protection against the digital misuse of a person’s voice, image, etc.

Read more about the Rights Alliance’s recommendations for enforcing against digital imitations, and our comments on the recently concluded consultation on the bill here.

It is very easy to upload copyright-protected content to AI services or to generate output that infringes the rights of others. The Rights Alliance has highlighted this issue by focusing on problems with customized chatbots based on study books here.

There is much inspiration to be found in the current rules of Article 17 of the DSM Directive, which require online content-sharing services to ensure that users cannot unlawfully upload protected content.

The expert group recommends that AI providers at the EU level be required to implement similarly effective safeguards that prevent users from uploading copies of copyright-protected works and thereby infringing copyright. Likewise, services should ensure that users cannot generate output that infringes copyright.

How liability obligations should be designed in practice can be determined through collaboration between AI providers and rights holders in each country – as is already the case in the Danish Article 17 Forum, where the Rights Alliance represents rights holders.

The expert group considers that there is a risk of artificially generated content replacing human-created content. It therefore recommends examining the risk that AI-generated content could outcompete human-made content, and how, in that case, the conditions for human-created content can be strengthened.

At the same time, various mitigation mechanisms could be explored at the European level, such as ensuring remuneration for the commercial use of content, regardless of whether it is artificially generated.

The expert group emphasizes that any such schemes must take into account the need to avoid unnecessarily restricting technological innovation or the competitiveness of Danish and European enterprises.

The expert group considers that there is a risk of artificially generated content replacing human-created content. It therefore recommends examining the risk that AI-generated content could outcompete human-made content, and how, in that case, the conditions for human-created content can be strengthened.

At the same time, various mitigation mechanisms could be explored at the European level, such as ensuring remuneration for the commercial use of content, regardless of whether it is artificially generated.

The expert group stresses that any such schemes must take into account the need not to unnecessarily restrict technological innovation or the competitiveness of Danish and European enterprises.

Respect for copyright in relation to artificial intelligence must be strengthened through awareness initiatives.

There is considerable uncertainty about the rules for using copyright-protected content in connection with AI. The expert group therefore believes that various awareness initiatives should be launched to guide companies and institutions developing artificial intelligence, as well as users of such services.

These initiatives should include official guidance from the authorities on the basic rules of copyright in relation to AI. In addition, funding should be allocated to develop educational initiatives, awareness efforts, and campaigns in collaboration with private actors – targeting institutions, students, and other AI users.

The Rights Alliance believes that valuable lessons can be drawn from Share With Care, a partnership between the Ministry of Culture, Teleindustrien, Dansk Industri, Dansk Erhverv and the Rights Alliance, which aims to promote the use of legal content

Rights holders must be given a stronger basis for enforcing their rights and negotiating agreements with AI providers, by clarifying in copyright law that the content AI services are trained on constitutes a communication to the public when the services are made available on the European market.

The use of copyright-protected works represents enormous economic value for AI services, but under current conditions providers have no incentive to enter into agreements with rights holders regarding their use. The majority of the expert group therefore believes that copyright law should be amended to specify that the content AI services are trained on is communicated to the public when the model is released on the market.

This would create a clear incentive to conclude agreements on such use, as the expert group is of the view that all major commercial AI services have been trained on vast amounts of protected content without the rights holders’ permission.