Russian artificial intelligence developers may face a de facto mandatory fee for training neural networks on publicly available materials. This risk was identified by the Association of Computer and Information Technology Enterprises (APKIT), commenting on the proposed amendments to AI legislation and the Civil Code.
The essence of the dispute is whether developers should pay copyright holders every time their texts, images, music, or other open content are used to train models. One of the discussed options involves collective management of such rights – based on the principle of mandatory deductions, which are then distributed among authors. Previously, a mechanism similar to the existing 'blank media tax' was even proposed in this regard.
APKIT considers such a scheme technically complex and warns that it could increase the cost of creating AI systems. According to the association's assessment, the additional financial burden could lead to a reduction in the number of projects and intensify competition for specialists with foreign companies.
At the same time, the association does not reject the idea of protecting authors. APKIT supports a mechanism whereby a copyright holder can prohibit the use of their materials for training neural networks in advance. A similar approach is already being discussed in the working group: the content owner will be able to publicly declare their refusal to allow its use.
The amendments are currently under discussion. The working group under the Codification Council is preparing proposals on how to combine AI development with copyright protection. In particular, the recognition of model training as a separate way of using works and rules allowing authors to control the inclusion of their content in training datasets are being discussed.