Technology
Putin's "sovereign AI" problem: you can't train a mind you keep blindfolded
The Kremlin promises a homegrown rival to ChatGPT, built on data its own censors keep shrinking.
![Russia's President Vladimir Putin (L) visits an exhibition of the AI (Artificial intelligence) Journey international AI conference in Moscow on November 19, 2025. [Kristina Kormilitsyna/POOL/AFP]](/gc6/images/2026/07/13/56901-afp__20251119__84j67we__v3__highres__russiapoliticsscienceairobot-370_237.webp)
By Olha Chepil |
Russia wants to build an artificial intelligence (AI) that thinks for itself. First, it is making sure the machine cannot read its own country's history.
Russia's internet regulator, Roskomnadzor, keeps blocking one foreign service after another, and Russians increasingly ask why they can no longer reach ChatGPT. President Vladimir Putin has an answer: the country will build its own. But the technology works against him. A model learns from the text it is fed, and Russia's censors have already trimmed the menu.
Speaking in late May at the Eurasian Economic Union summit in Astana, Putin claimed Russia stands among only three countries able to develop sovereign AI. He cited energy, talent and intellectual potential as its advantages. Days later, at the Saint Petersburg International Economic Forum, he named AI, autonomous systems and platform solutions the core technologies of the future and said Russia held "strong prospects."
Pro-war Z-communities answered fast. One verdict spread widely: "Not a technological breakthrough, but a leap into the void."
![Russia's President Vladimir Putin (C) chairs a meeting on the development of artificial intelligence technologies in Moscow on April 10, 2026. [Alexander Kazakov/POOL/AFP]](/gc6/images/2026/07/13/56900-afp__20260411__a7fx4g3__v1__highres__russiapoliticstechnology-370_237.webp)
A whitelisted sovereignty
While Putin talked up global leadership, Roskomnadzor moved the other way. By late February 2026, Russia had blocked nearly 500 virtual private network (VPN) services, a 70% jump in three months. VPNs remain legal on paper. In practice, they barely work.
Demand climbs anyway. The share of Russians using VPNs rose from 4% in 2021 to 42% in 2025. In December 2025 alone, Russians ran more than 12 million Yandex searches for "VPN," an all-time high.
Software engineer Sergejs Roze, a Latvian citizen who tracks the Russian tech sector, sees the effect when he messages people inside the country. Acquaintances vanish. Some scrub their links to anyone the government dislikes. Others log on once a day in the morning, while a connection still holds.
"This is clearly tied to attempts to regulate the internet so people don't read 'bad' things there," Roze told Kontur.
Oleksandr Krakovetskyi, who heads the software firm SeoDevRein and has 15 years in the industry, warned against underestimating Russia. Before the war, Yandex competed with Google, and banks such as Sberbank and Tinkoff poured money into digital innovation. The country built domestic copies of nearly every major global product, he said. They were clunkier, but the market was real.
The war upended that. Developers fled, and sanctions choked off Western microchips. The drive toward isolation is deliberate and gaining speed, Krakovetskyi told Kontur.
"Sovereignty means having total control over infrastructure, data, and information flows," he said.
Russia is copying a familiar model, he added. Iran can switch off its internet at will. North Korea sits almost entirely offline. Russia is assembling the same kind of system, piece by piece.
Filtered AI
The hardest problem for Russian AI may not be money or computing power. It is data.
Large language models train on enormous text collections, and a broader, more varied base yields a sharper result. Isolation guts that advantage. Roze expects Russian AI to learn only from material that clears the censors.
"You can feed it Soviet propaganda, but 'The Gulag Archipelago' is off-limits," he said. "The smaller the database, the worse the output. It will simply perform poorly."
One shortcut exists. Russia can take an open-source Western or Chinese model and bolt a censorship filter on top. Roze said that is the likely route: let the foreign system do the heavy lifting, with a small Russian layer policing what it says. Sberbank and Yandex have tried versions already. When their tools produced images of Putin missing his "wings and a white horse," the results were quickly "corrected."
The money gap is stark. Russia set aside about 10 billion rubles ($125 million to $130 million) for its "Artificial Intelligence" federal project in 2026. Meta alone may spend up to $135 billion on AI this year. The gap runs into the hundreds.
Controlling the dataset also means controlling memory, said Dmytro Hromakov, a sociologist and communications expert at the International Center for Countering Russian Aggression. Open the system to outside sources, he told Kontur, and the AI might revisit history and offer a more honest picture. For a totalitarian regime, he said, that is unacceptable.
A quiet adaptation
Most Russians are unlikely to treat sovereign AI as a threat. After the full-scale invasion, many of the most active and critical citizens left, Hromakov said. Those who stayed press the authorities less.
Russia, he argued, is becoming a nation of "top managers" — people who run what others built rather than create it themselves.
"They are people-functions. And a function cannot question the will of its creator," he said.
In that frame, blocked services, the fading of YouTube and controlled AI register as nuisances, not politics. People adjust to their conditions rather than fight them, he said. That resilience is the point. Putin's approval is slipping, but slowly. The frustration is real, yet it stops short of protest.
"We shouldn't expect rapid change," Hromakov said. "These processes are built for the long haul."