WARSAW—Russia is trying to turn its technological disadvantage into political leverage abroad. On September 1, Russia’s first federal artificial intelligence (AI) law will take effect, putting a legal framework around what Moscow calls “sovereign AI.” The law establishes technological independence, state support for Russian AI developers, and overseas promotion as national goals. Additional regulations will follow, while stricter rules defining which large AI models qualify as “sovereign” or “national” will take effect in March 2027.
While Russia’s AI ecosystem lags behind those of the United States and China, Moscow can pitch its systems as giving governments more latitude to shape outputs around local information and government-defined political preferences—and less reliance on Western tech companies.
The key issue is what kind of sovereignty Russia’s customers will actually get. A related question is whether authoritarian buyers might use these systems to further political repression.
Bringing the model into the censorship system
The new AI law, which Russian President Vladimir Putin signed on July 26, applies to models with at least one billion parameters—the values a model learns during training, which guide how it generates answers. While this law does not cover the country’s entire AI ecosystem, it does include many large models that are politically and economically strategic.
The law creates two exclusive categories: “sovereign” and “national” models. Both must generate responses and store data in Russia-owned data centers located in Russia. A sovereign model must stay under the control of a Russian developer throughout its life cycle and be fully reproducible. A national model has more leeway: It can use foreign open-license components as long as they remain under the Russian developer’s control.
Both categories must also pass a government process that checks whether they comply with Russian law and “traditional Russian spiritual and moral values.” For years, the Kremlin has used this language to describe its conflict with the West, to justify repression and censorship in Russia, and to support its version of the war against Ukraine. In 2022, a presidential decree named the United States and other “unfriendly countries” as threats to these traditional values.
The law does not itself provide developers with a blacklist of forbidden prompts or dictate exactly how models should answer questions on topics sensitive to the Kremlin. Instead, its main influence comes earlier in the process. Developers who want their models to be recognized as sovereign or national must show the state that their models follow Russia’s existing speech restrictions and ideological standards.
This may incentivize developers to comply in advance and even go beyond what is strictly required. Russian law already restricts speech about the war in Ukraine, “extremism,” “foreign agents,” and other sensitive topics. Developers may lean toward more conservative guardrails to ensure compliance: for instance, refusing certain requests, omitting disputed information, or defaulting to official terminology and narratives.
There is already evidence of restrictive behavior in Russian models. A 2025 study found that Russian models, such as GigaChat and YandexGPT, show high levels of “hard censorship”—refusals to answer queries, rather than biased or incomplete responses.
Russian officials have been direct about such regulation. In April, Deputy Minister of Digital Development Alexander Shoitov, whose ministry proposed and drafted the law, said that AI should have censorship mechanisms at both the input and output stages that block certain user questions.
Sovereignty on Moscow’s terms
Russia’s legal definition of sovereign AI is intended to achieve something its technology sector cannot yet deliver: its own independent AI supply chain. This gap is visible across research, computing power, talent, and access to AI chips.
In Stanford’s 2024 Global AI Vibrancy Ranking, Russia was ranked twenty-eighth out of thirty-six countries when it came to research, talent, investment, infrastructure, and policy, far behind the United States and China. Russian-language AI models still trail leading models. Russia spent about 0.94 percent of its GDP on research and development in 2023, while China spent 2.6 percent. The World Intellectual Property Organization ranks Russia twenty-eighth in human capital and research but sixty-second in knowledge and technology outputs. Scientific emigration and cuts to civilian research since Russia’s full-scale invasion of Ukraine have further weakened its research base, making it harder to compete at the technological frontier.
Access to the hardware needed to train the models has also narrowed. Late last year, the Wall Street Journal reported that Russia’s imports of graphics processors and other AI-related chips dropped by 84 percent in 2024 compared to the levels before the 2022 war. Even though this figure does not include some shipments made through third countries, it does indicate the effect that sanctions and export controls are having on Russia’s technology sector.
That has pushed Russia toward a new foreign dependency. In May 2026, Sberbank, the state-controlled bank leading Russia’s AI development effort, said it was looking for Chinese chips to power GigaChat.
The new AI law recognizes the gap between Russia’s ambitions and its capabilities by letting national models use foreign components released under open licenses. Russia’s strategy therefore combines foreign technological inputs with Russian political and commercial control. But the strategy contains a paradox: Moscow plans to export the promise of technological independence using systems that rely on foreign components and a supply chain it cannot control.
Turning domestic control into an export offer
The export ambition is written directly into the new law. Article 2 lists overseas promotion as one of its aims. Article 7 makes it clear that supported developers can export and do joint research. Article 12 makes promoting Russian AI models abroad part of the country’s international AI policy.
Sberbank started promoting this idea even before the law was passed. In June, First Deputy CEO Alexander Vedyakhin told Reuters that Sberbank was focusing on countries in Africa, Asia, Latin America, and Oceania that want their own AI but cannot afford to build it themselves. Vedyakhin admitted to a trade-off: The first systems might be slower and less advanced than top foreign models, but they are more affordable, can be adapted locally, and meet state needs. The BRICS group’s AI Alliance Network and Russia’s bilateral ties could help Moscow connect with these buyers.
However, the law leaves a basic question unanswered: What would a foreign government actually own or control? Does a government buying Russian AI get copies of the model that it can operate, modify independently, and run on local infrastructure? Or can it only access a Russia-based service? Who manages the updates, content filters, audit access, customer data, and decisions about suspending service?
Relying on Russian models could give Moscow influence even if there is no hidden malicious capability in their models. Russian providers work under laws that give the state extensive powers of surveillance and access to data. If a model is hosted or managed from Russia, this legal environment could affect foreign users through control retained by the provider. Dependence can also grow through security updates, technical support, and other services. The law does not guarantee that foreign operations or customer data will be protected from Russian jurisdiction. The result could be that “sovereign AI” shifts technological dependence rather than eliminating it.
Nor would authoritarian practices necessarily come just from Moscow. The government buying the system could also demand political filtering, surveillance, or monitoring. For example, an authoritarian buyer could require the system to suppress opposition content, label protests as security threats, or use its official terminology to answer political or historical questions. If used in government systems, it could also track sensitive questions or flag users for extra scrutiny.
Closing an information gate, opening an export market
Russia’s sovereign AI efforts did not emerge in 2026. Moscow introduced its first national AI strategy in 2019 and then updated it in February 2024, after its full-scale invasion of Ukraine and increased Western sanctions deepened its technological isolation.
The new law comes as several pressures converge. Generative AI is becoming a front door to information and services. Russia’s technology gap with the United States and China is deepening. Sanctions are restricting Russia’s access to Western technology and it is relying on Chinese hardware to meet demand. Moscow is acting now to prevent foreign AI models from taking hold at home and to get ahead of competitors in emerging markets.