By Andrew Osborn

March 20 (Reuters) - Foreign AI tools like Claude, ChatGPT and Gemini could be banned or restricted inside Russia if they fail to adhere to new rules that would give Moscow sweeping powers to regulate the sector, according to government proposals published online.

The proposals, published by Russia's Ministry for Digital Development, would extend to the burgeoning AI sector Russia's drive to establish a sovereign internet - protected from foreign influence and respecting what it calls "traditional Russian spiritual and moral values."

Russia's Ministry for Digital Development said in a statement that the new rules were designed to "help protect citizens from covert manipulation and discriminatory algorithms."

RESTRICTING CROSS-BORDER AI TECHNOLOGY

The initiative, which is likely to benefit home-grown AI tools being developed by state lender Sberbank and technology group Yandex, has been made public at a time when the Russian state is tightening state control over the internet.

The regulations are expected to enter into force next year after further review and government approval.

"The operation of cross-border artificial intelligence technologies may be prohibited or restricted in cases specified by the legislation of the Russian Federation," the rules state.

The state-run RIA news agency reported on Friday that foreign AI tools would fall under the new rules because they inevitably transferred the data of Russian citizens abroad.

"Cross-border artificial intelligence technologies refers to all foreign AI models, including ChatGPT, Claude and Gemini, where the use of such models results in user data, queries and dialogues being transmitted to the developers of these models outside Russia," RIA cited a specialised technology lawyer, Kirill Dyakov, as saying.

All three models mentioned by Dyakov were developed by U.S. companies: OpenAI, Anthropic and Alphabet's Google GOOGL.O, respectively.

Other foreign but open AI tools, such as China's Qwen or DeepSeek, could however be safely adapted and rolled out in a closed environment on the proprietary infrastructure of Russian government organisations and companies, Dyakov said, since any data processed would remain within that infrastructure.

RIA said AI models used by more than 500,000 people per day would need to store Russian user information on Russian territory for three years to be compliant under the new regulatory regime. Western tech companies have in the past refused to comply with such demands.

(Reporting by Andrew OsbornEditing by Tomasz Janowski)