The Russian Government Commission has approved a draft law on AI regulation, focusing on sovereignty and national security.
On June 22, the government commission on legislative activities gave the green light to a document aimed at supporting the development of artificial intelligence in Russia. According to official sources, the key element of the bill is a clear division of large foundational AI models into two categories: "sovereign" and "national."
Two Categories: Sovereignty vs. Open Source
The new regulation applies only to models with more than 1 billion parameters. "Sovereign" models are those fully developed by a Russian legal entity at all stages, using exclusively local infrastructure. "National" models, in turn, may partially rely on open-source components.
Deputy Prime Minister Dmitry Grigorenko emphasized that these two categories will receive priority state support. Plans are underway to implement such solutions in the most sensitive areas, primarily in public administration. This is a logical step, given the global trend toward digital sovereignty: control over data and algorithms is becoming a matter of national security.
Abandonment of "Trusted" Models and Labeling
Notably, the category of "trusted" models for critical information infrastructure has disappeared from the final version of the bill. As informed sources explained, this idea was abandoned due to already stringent requirements from the FSTEC and FSB — duplicate regulation would have been redundant.
Also excluded from the document were a number of controversial and potentially growth-stifling norms: mandatory labeling of AI-generated content, copyright blocks, data center regulation, and restrictions on foreign neural networks. Grigorenko directly stated that the project does not provide for a ban on foreign solutions — this is a pragmatic approach that allows maintaining access to the best global practices.
Timelines and Transition Period
The bill is expected to be submitted to the State Duma by the end of this week. The main provisions are set to take effect on September 1, 2026. Norms concerning the government's authority to define models and developers' obligations will come into force on March 1, 2027.
For already deployed models that do not meet the new criteria, a lenient transition period is provided until September 1, 2032 — provided that data processing is carried out on the territory of the Russian Federation. This gives businesses sufficient time to adapt.
Analytical Commentary: This bill is not just technical regulation but a strategic maneuver. The rejection of excessive requirements (labeling, restrictions on foreign models) indicates an understanding that the AI market is in a phase of active growth, and excessive pressure could stifle innovation. The focus on "sovereign" and "national" models is an attempt to create local champions capable of competing with global players while remaining under full state control. The question is whether Russian infrastructure can provide the necessary computing power for such models without relying on foreign chips.