The State Duma has completed work on a fundamental document that will change the landscape of the digital market in Russia. In the second and third readings, a law on state regulation of artificial intelligence was adopted. The document not only introduces new terms into the legal field but also lays down a rigid architecture for the development of domestic neural networks, dividing them into strictly hierarchical categories.
Sovereign vs. National: The New Classification
The central element of the adopted law is the division of AI models into two key types: 'sovereign' and 'national'. This distinction determines not only technical implementation but also the strategic role of technologies in the country's economy.
To obtain the status of a sovereign model, a developer must meet strict criteria: the system must be created entirely in Russia and use exclusively Russian hardware and software components. According to RBC sources on the IT market, currently, only Sber's development fits these criteria.
The category of national models allows the use of foreign components, making it more flexible for existing ecosystems. Nevertheless, a single strict rule applies to both types of models: servers and databases must be located exclusively within the territory of the Russian Federation. According to current data, Yandex's technology may fall into this category.
Values and Security: Principles of Regulation
The law establishes the basic principles of AI development in the country. These include technological independence, security, and the protection of human rights. However, special attention is drawn to the clause regarding the consideration of 'traditional Russian spiritual and moral values'.
This aspect sparked sharp discussions in the business community even at the stage of publishing the first version of the draft law by the Ministry of Digital Development in March. Despite concerns from industry representatives about how these abstract concepts would be translated into code, the final text of the law retained this wording unchanged.
Legal Disputes: Copyright and Labeling
One of the most controversial points of the document was the norm on copyright. Russian legislation allows the use of materials protected by copyright but available in the public domain for training AI models. The key condition is the absence of the need to obtain the consent of the copyright holder. This approach, which causes disputes on the international stage, has been chosen as a priority to stimulate the development of domestic algorithms.
Separate requirements will affect owners of popular platforms. Social networks, media outlets, and marketplaces with a daily audience of more than 500,000 people will be required to provide users with tools for labeling content created with the help of AI. The labeling itself will remain voluntary, but the obligation to provide such a function will come into force on March 1, 2027.
Politics of Compromise: What Was Removed from the Law
The final text of the document was the result of tense negotiations between regulators and business. The initial version of the draft law included a third category — 'trusted' AI models. The requirements for them were even stricter than for sovereign ones and implied full control at all stages of the life cycle.
Major Russian companies and industry associations criticized this initiative sharply, warning of risks for the development of the industry. As a result, market pressure worked: the category of 'trusted' models was removed from the final text. Nevertheless, the government will receive broad powers: it will be able to determine in which cases the use of only sovereign models is allowed and to establish measures of state support, including access to training data.
The law will come into force on September 1, 2026, subject to its signing by President Vladimir Putin. This date will become the starting point for a new era of digital sovereignty in Russia.
