The Ministry of Digital Development has published a draft federal law on state regulation of artificial intelligence on the portal of regulatory legal acts. The law is planned to come into force on September 1, 2027. The key innovation is the concept of sovereign, national, and trusted AI models. Sovereign and national models must be fully created in Russia by citizens and legal entities from the Russian Federation based on data from Russian sources. Trusted models are entered into a special register. For critical industries — medicine, critical infrastructure, public administration — the use of foreign models is planned to be limited.
Labeling and responsibility for AI content
All content created by neural networks is subject to mandatory labeling in a format that is human-readable and machine-readable. The labeling is provided by the owner of the AI service. Large social networks are obliged to check for labeling: if it is absent, they must independently label the material or delete it. The use of AI in committing crimes — creating deepfakes, interfering in elections, manipulating public opinion — will become an aggravating circumstance.
Responsibility along the chain: developer, operator, user
Developers are obliged to exclude discriminatory algorithms and block the creation of illegal content. Operators must inform users about interaction with AI. Users are responsible for distributing generated content. The law does not apply in the areas of defense, state security, and emergency situations.
The government has chosen a soft approach: it regulates not technologies, but products and relationships in their application. The authorities fear that an excess of restrictions on the Russian market — about 2% of the world market — will push companies and specialists to other jurisdictions. Russia is now among the most liberal countries in the field of AI development: the EU and China already have strict registers of models and data localization requirements, which has caused an outflow of specialists from these markets.
Key parameters of the draft law
- Entry into force: September 1, 2027
- New concepts: sovereign, national, trusted AI model
- Sovereign model: created in Russia, on Russian data, by Russian legal entities
- Register: of trusted AI for state systems and critical infrastructure
- Labeling: all AI content, the responsibility of the service owner
- Social networks: check labeling, if absent — label or delete
- Aggravating circumstance: the use of AI in crimes
- Exceptions: defense, state security, emergency situations, law and order
Read more materials on the topic:
- ChatGPT, Claude and Gemini: Russia is preparing restrictions for foreign neural networks
- The right to human contact: Russia wants to enshrine the possibility of refusing AI service
- AI fakes will not pass: a service for labeling content created with the help of neural networks is being created in Russia