Government Policy in Artificial Intelligence: Easing Regulations and Strengthening Digital Sovereignty

The updated version removes restrictions on the use of data for training AI models: developers are now allowed to use any available datasets.

The Draft Law introduces three categories of AI models: sovereign, national, and trusted. The latter are intended for use in government information systems and critical infrastructure facilities. They must undergo security certification, operate within Russian jurisdiction, and confirm their quality with relevant regulators.

The initial version of the document required that models seeking “sovereign” or “national” status be trained exclusively on Russian data and developed by Russian citizens. Market participants pointed to the insufficient volume of relevant Russian-language data, which could have led to a deterioration in system quality. Following criticism, these provisions were removed.

During the discussions, industry representatives noted the absence of fully sovereign models in Russia, as most solutions rely on foreign components and open datasets. This was followed by proposals to ease the requirements.

Work on the document continues at the Digital Economy Autonomous Non-Profit Organization platform. The classification of models is aimed at supporting domestic developers and reducing technological dependence. National models are allowed to use open-source software, while detailed criteria will be established in subordinate regulations. Confirmation of trusted status is required only for solutions used in the public sector. Foreign AI systems may operate in Russia in compliance with national legislation and may qualify for trusted status if they meet information security requirements.

Author: Candidate of Economic Sciences, Associate Professor of the Department of World Economy and World Finance, Financial University under the Government of the Russian Federation  Alexey Alekseevich Soluyanov.