Russia’s AI Law Tilts State Funding Toward Domestic Models Without Banning Foreign Tools

Published by James Harris on June 23, 2026June 23, 2026

Russia’s government has approved a draft AI law that creates the country’s first legal framework for large language models, dividing them into “sovereign” and “national” categories based on how domestically they are built and operated. The legislation does not ban foreign AI tools but makes clear that Russian public institutions will primarily run Russian-built models, with state support flowing accordingly.

The bill’s structure reflects a pattern becoming familiar across non-Western markets: regulatory frameworks that stop short of outright prohibition while systematically tilting economic incentives toward domestic alternatives. Japan’s Sakana AI explicitly cited the gap created by US export controls on top American AI models as a direct rationale for building local capacity, and Russia’s legislation operates on similar logic. The difference is that Russia is encoding the preference into law rather than leaving it to market dynamics. The bill covers models with more than one billion parameters, splits them into sovereign (fully domestic infrastructure) and national (domestic-led but may use open-source components) tiers, and dropped an earlier third category of “trusted” models after industry feedback. Several provisions that would have created real compliance burden, including mandatory labeling of AI-generated content and specific accountability rules for misuse, have been softened or removed entirely.

Thirteen articles and thirteen pages is a remarkably thin foundation for regulating an entire technology sector, which suggests this is a placeholder framework that will be filled in later.

The timing matters. Russia currently has no AI legal framework at all, and Deputy Prime Minister Dmitry Grigorenko said explicitly that laws are needed before significant decisions about the technology can be made or support mechanisms introduced. The State Duma could receive the bill by the end of this week, and the core provisions are set to take effect September 1, 2026, with agency-level rules following in March 2027. That 18-month runway before enforcement gives developers time to adapt but also gives the government time to reintroduce the provisions it just stripped out, including data center regulations and cross-border AI restrictions, through secondary legislation. The authorities said explicitly they are leaving that door open.

The practical effect for now is definitional: Russia has established what counts as a sovereign AI model and signaled that procurement and public-sector contracts will follow that definition. Companies building in the space now have a target to aim at, even if the incentive specifics are still being worked out.