Back to News
RSS feedtechpolicy.press

Russia’s AI Law Prioritizes Control Over Capability

Summary

Russia’s first comprehensive law governing large foundation models took effect on September 1, 2026. The article argues that the law is designed less to mitigate AI risks than to support and control the developers of models that meet Russian requirements. It applies to large foundation models with at least 1 billion parameters, including models made available to users in Russia, although that threshold can cover both relatively small local models and systems far beyond it. The law creates “sovereign” and “national” statuses. Both require a Russian legal entity, Russian-owned data centers in Russia, and conformity with Russian law and undefined “traditional Russian spiritual and moral values.” A sovereign model must be developed end to end by its Russian developer, while a national model may be based on a foreign open model if the developer changes its essential characteristics. This creates a legal route for companies such as Sber, Yandex, T-Bank and MTS to build on foreign foundations, especially Chinese models such as Qwen, while Western export controls restrict access to advanced hardware and some services. Status holders may receive state support, access to state datasets, a role in policy formation and preferential access to markets where the government requires a sovereign or national model. The law also places chatbot and assistant services under existing rules requiring interaction data, including message content, to be stored in Russia and made accessible to the FSB. From March 2027, it permits certain temporary copies of copyrighted material for training qualifying models, without establishing remuneration for rights holders, and lets users attach notices to AI-generated audio or visual material rather than requiring them. Many procedures remain dependent on future government decrees, including conformity testing, status decisions and access to state data. The law leaves AI uses in defense, public order, biometric identification and predictive policing largely to executive rulemaking, without specified prohibitions, independent oversight or procedures for affected people to challenge decisions. The authors conclude that Russia is using model status to combine technological sovereignty, industrial support and control over downstream markets: sovereignty concerns jurisdiction and domestic control more than frontier capability.