The draft law on artificial intelligence regulation was significantly rewritten before being considered by the government. The document was reduced to 13 pages and 13 articles, and its focus shifted from broad market control to supporting major Russian AI developments.
Only large fundamental models with more than a billion parameters will fall under the new rules. Small solutions, including neural networks based on open source code, will not be affected by the regulation.
The project retains two statuses – “sovereign” and “national” model. The first must be entirely created by a Russian legal entity and operate on infrastructure in Russia. For the second, the use of open-source components is allowed, but the key development must also be Russian. Only such models will be able to claim state support and be used on critical infrastructure.
At the same time, many of the most controversial norms have been removed from the text. Mandatory labeling of AI content is no longer provided: developers and platforms will only have to give users the opportunity to apply it. Issues of copyright, training on protected data, responsibility for the results of neural networks, regulation of data centers, and cross-border models have been moved to future subordinate acts.
The main provisions of the law are planned to be launched on September 1, 2026, and the government's powers regarding the application of models and developers' obligations – from March 1, 2027. Systems already in operation that do not fall under the new criteria will be given a transitional period until September 1, 2032 – provided that data is processed in Russia.
Our task is to create conditions under which domestic AI developments will receive priority support and be more actively implemented in key sectors of the economy.
However, this does not mean a ban or restriction of access to foreign neural networks. The bill may be submitted to the State Duma by the end of the week.
