A law governing large language models has come into effect in Russia in September 1st, 2026. While many countries are prioritising AI governance, Russia's law highlights the importance of AI foundation models that would strictly conform with Russian law and what the statute calls “traditional Russian spiritual and moral values.”
The Russian state is known for being heavily involved in the digital space of the country. This AI law, however, is narrow in focus in that it does not offer a comprehensive guide to AI systems overall, but only regulates large foundation models that have over 1 billion parameters and 'are available to users in Russia'. This wording can include both small open models and also any foreign developers that serve the Russian market.
AI models are categorised into 'sovereign' and 'national', and both face the same requirements: the developer must be a Russian legal entity; responses to user queries must be generated and stored in Russian-owned data centres in Russia; the model must be assessed for conformity with Russian legislation and adhere to the moral and spiritual values of the country. The law does not state, however, how assessment will be made to ensure the models meet the stated requirements.
A 'sovereign' model must be developed end-to-end by the Russian developer. A national model may be built on a foreign model under an open license, provided the Russian developer determines and changes its essential characteristics. Some of the key AI models being built in Russia rely on Chinese foundations such as QWEN or DeepSeek, since the major Western providers block Russian IPs. The 'national' model status can thus be interpreted as a legal pathway for Russian developers to continue building on open foreign models, primarily Chinese models, as the AI development of native Russian models is relatively behind the top commercial AI models available worldwide.
