Russia revises AI regulation bill — ‘sovereign models’ remain but can now be trained on non-Russian data

Russia’s government has circulated a revised version of its artificial intelligence regulation bill — a document that has been cut nearly in half, from 21 articles to 13, the Russian business news outlet RBC, Izvestia, and the Russian business daily Kommersant reported.

The bill has been renamed “On Supporting the Development of AI Technologies,” replacing its previous title, “On the Fundamentals of State Regulation in the Areas of Application of AI Technologies.” The revised document no longer addresses artificial intelligence broadly; it focuses instead on the development, deployment, and use of large foundation AI models.

The new version also drops the concept of a “trusted” AI model — a designation authorities had previously planned to apply to systems approved for use in government and at critical information infrastructure facilities. Only “sovereign and national models” remain. That status will be granted to models developed by Russian legal entities, provided they process user requests and store data in Russia and comply with Russian law and moral and spiritual values.

A source at RBC familiar with the IT market said the Russian state bank Sberbank’s model meets the criteria for a sovereign large foundation model, while the Russian technology company Yandex’s model qualifies as a national one.

The revised bill also removes the requirement that models seeking “sovereign” or “national” status be trained exclusively on data of Russian origin or developed solely by Russian citizens.

Izvestia reports that the new version contains no provisions allowing the government to ban foreign neural networks in Russia. Deputy Prime Minister Dmitry Grigorneko said businesses would retain the right to independently choose the technological solutions that best serve their needs. Experts who spoke to the outlet said many Russian AI services are built on open-source foreign models — such as DeepSeek — and that a ban would have shut down working products.

The first version of the bill was criticized by Russian business representatives. They argued that if passed, companies’ costs for implementing AI would rise, product launches would slow, development would move to other jurisdictions, and Russians’ access to advanced technologies would be restricted.

Experts raised the most objections to the bill’s requirements for “sovereign” and “national” AI models, whose development and training stages must take place in Russia, on Russian datasets, and under the guidance of Russian citizens and companies. Companies noted that Russia currently has no models meeting all the stated criteria for “sovereignty,” as all Russian developments use foreign components and open datasets.