The Federation Council approved a framework law on July 17 regulating the development, implementation, and use of artificial intelligence systems. This document introduces the key industry concepts of "sovereign" and "national" AI models, and also establishes measures for state support of developers and rules for labeling content created by neural networks.
Sovereign vs. National Models: What Changes?
Lawmakers have clearly divided large fundamental models (LFMs) — systems with a billion or more parameters capable of solving complex tasks at a human level. Two types of such models will now operate in Russia:
| Model Type | Developer and Control | Infrastructure | Foreign Components |
| Sovereign | Russian legal entity with full control at all stages | Only in domestic data centers | Completely excluded |
| National | Russian legal entity | Domestic infrastructure | Allowed under open license terms |
Both types of models must comply with Russian legislation and traditional spiritual and moral values. In return, developers gain access to financial, property, and informational support from the state. A key point is the legal opportunity to train neural networks on data from federal and regional information systems.
Content Labeling and a General Shift Toward State AI
A separate provision requires owners of websites, social networks, and applications with a daily audience of more than 500,000 users to implement a tool for labeling materials created by neural networks. The format of such labeling will be developed through an agreement between service developers and consumers.
These changes are part of a general trend toward strengthening state control in the technology sector. Simultaneously, the Ministry of Digital Development is designing national platforms for managing communication networks, and the FAS proposes recognizing the circulation of AI technologies as a strategic activity. However, the RATEK association sharply opposed the latter initiative, warning of a risk of electronics shortages due to vague wording.
Disputes are also brewing in the industry over copyright issues when training neural networks, and deputies are discussing a possible restriction on access for children under 14 to social networks and chatbots in 2027-2028. Notably, these restrictions are planned to be introduced simultaneously with the launch of the school profile "Artificial Intelligence" starting September 1, 2026.
Expert Opinion: The adoption of the framework law is an important step, but it only sets the outlines. Key issues — access to data for training, copyright protection, and clarity of wording for businesses — remain open. The market expects not so much regulation as predictable rules of the game that will not stifle innovation with excessive bureaucratic barriers.