Russian AI developers may be obliged to pay copyright holders for works used to train neural networks. This applies not only to new models: the draft amendments provide for a one-time compensation for already completed training. The procedure and amount of payments are to be determined by the government.
The authors of the initiative propose to enshrine in the Civil Code that AI training itself is considered the use of objects of copyright and related rights. The copyright holder will be able to publicly prohibit the use of their content for such purposes — for example, by placing a notice on the website, next to the work, or on a special state resource.
If there is no prohibition, the material will be allowed for use, but free of charge — only with the author's separate consent. In other cases, the developer will have to pay remuneration. The new rules may apply not only to national and sovereign models, but generally to companies and other individuals who train or retrain AI.
The amendments were prepared by the working group of the Presidential Council for the Codification of Civil Legislation. They are planned to be discussed on August 19, and a broader package of proposals for regulating AI and copyright is expected by the end of September.
The dispute over training neural networks on third-party content has been ongoing for a long time. Copyright holders consider the current exceptions too broad, and according to HSE University estimates, their total losses due to generative AI in 2025–2030 could reach 1 trillion rubles.