Copyright-related provisions have only been part of the debate on the AI Act since the European Parliament's proposal in June 2023. The European Parliament proposes transparency obligations for providers of foundation models with regard to training data protected under copyright. These obligations are intended to support the enforcement of existing copyright provisions in the context of foundation models.
Bitkom continues to take the position that copyright provisions should not be part of the AI Act at all. Should the trilogue negotiations result in the inclusion of specific provisions on copyright, Bitkom takes the opportunity to raise serious concerns about the existing proposals and to propose a practical and proportionate solution to this highly relevant and fundamental issue.
Bitkom considers the proposed provisions as technically infeasible, a barrier to innovation, overly broad, a risk for trade secrets as well as proprietary information and of doubtful usefulness. Rightholders’ interests in transparency and control over the use of their works for AI training are well met by the existing copyright framework and technical measures that are already widely used. Therefore, Bitkom proposes to delete the proposed obligations. Instead, and more closely in line with the stated objective of supporting enforcement of existing copyright provisions, Bitkom suggests, that within the scope of Art. 28b providers shall be required to report on how they comply with existing technical regulation, in particular regarding opt-outs expressed by rightholders in line with Art. 4 of Directive 2019/790/EU.