The main danger of AI in authoritarian regimes is the ability to suppress any protest
Originally published on Global Voices
Moscow subway cameras are connected to a facial recognition system. Screenshot taken from the YouTube video “Facial recognition cameras help detain dissidents in the Moscow metro” by the DW YouTube channel; fair use.
Sonya, from Stavropol Krai, Russia, is 15 years old. In June 2026, she was sentenced to 70 hours of compulsory labor, connected to a February 2026 incident in which she had written the surname “Navalny” on a bus stop. That stop happened to be in a district that was being monitored by officers from the Center for Combating Extremism, commonly called Center E. Similarly, Yuri, a pensioner from Astrakhan who was also identified from surveillance footage, was given three years of restricted freedom for writing “No to War” in red marker on buildings.
Authoritarian regimes are increasingly using artificial intelligence (AI) as a technology of political survival. Cameras with facial recognition, video analytics, biometric databases, digital IDs, social media monitoring, Deep Packet Inspection (DPI) systems, and public service platforms are gradually merging into a single infrastructure of control.
Such a system makes it possible not only to observe citizens in real time, but also to reconstruct their past movements, connections, habits, communications, and political activity. The main risk is that AI and related technologies make repression faster and less visible: the state no longer needs to constantly demonstrate force if citizens know they can be identified retroactively.
Modern digital authoritarianism is not developing in isolation, but through the international exchange of technologies, standards, and practices. China, Russia, Iran, Belarus, Venezuela, countries in Africa, Western and Central Asia, and companies connected to them, sell and exchange AI infrastructure tools that are being used for the purpose of repression. Western technologies are also being bought and exchanged within authoritarian regimes.
Large language models
In early June 2026, Sberbank Deputy Chairman Alexander Vedyakhin told Reuters that the company was offering countries in the Global South access to Russian AI models that would take “local values” into account. Vedyakhin was referring to specific large language models (LLMs) — GigaChat and YandexGPT — which Russia had been trying to develop since 2019.
However, as IT expert Alexander Isavnin told journalists at the podcast Zhivoy Gvozd, Russian generative AI models lag behind American and Chinese models and cannot even generate text in Cyrillic. Yet, new national and sovereign AI models are supposed to reflect traditional values, including patriotism, service to the Fatherland and responsibility for its fate, a strong family, historical memory and continuity between generations, and unity of the peoples of Russia. These models are expected to be trained on domestic, authentically Russian material.
It is reasonable to assume that in countries like those in Western Asia, Russian AI models will reflect adherence to Islamic traditions, while in Argentina, perhaps, they will reflect respect for football from an early age. Developers, meanwhile, do not understand how people with no idea about how any of this works can write AI laws: “In places, the document reads as if it was written by people who had only ever seen LLMs in local news.” From the perspective of global democracy, however, this illiteracy among Russian lawmakers suggests that Russia is unlikely to supply LLMs to autocracies around the world.
China, by contrast — unlike American developers — makes its language models open in principle, meaning that anyone can take the code and modify it for their own needs. In China, all LLMs are controlled and tightly regulated by the state, so party censorship is written into the code from the start — for example, you cannot get an answer about the Tiananmen Square protests. However, the models are distributed as open source, which is why more than 50 percent of Chinese model users come from developers, including those in the Global South.
At the same time, the U.S.-based ChatGPT can easily replicate AI-generated Iranian propaganda videos, in which the main characters are Lego figures. Only once, a France 24 investigation showed, did the chatbot alert the journalist that he might be inadvertently reproducing propaganda, but desisted when he said it was only satire.
Mass surveillance systems
AI also includes other tech developments, like the so-called “smart city” systems which, in the Chinese version, imply comprehensive surveillance of a state’s own citizens. This can include video surveillance, wiretapping, internet censorship, facial recognition systems, ethnic profiling, and so on. In supplying such systems to the Global South, China has overtaken Russia.
Yet, Russia is not lagging behind in the use of AI to surveil its own citizens. State media write that Moscow has long moved beyond the threshold implied by the concept of a “smart city.” Today, it includes a complex digital urban ecosystem made up of video surveillance, AI, big data, and the “internet of things.”
According to estimates by Comparitech, two of the ten cities outside China with the highest number of cameras per person are Russian: Moscow and St. Petersburg. The list also includes four Indian cities, Seoul in South Korea, Baghdad in Iraq, Kabul in Afghanistan, and Singapore. In China itself, the number of cameras has practically reached a ratio of one camera for every two people: the largest mass-surveillance system in the world.
Russia has sold, and in some cases, continues to supply its surveillance software products to other authoritarian states, while remaining technologically dependent on China. In 2024, up to 80 percent of all cameras used in Russia were made in China and sold under Russian brands.
China has built one of the most sophisticated domestic systems of digital control; it has also turned the export of digital infrastructure into part of its foreign policy via the Digital Silk Road, launched back in 2017. Through the initiative, China promotes digital infrastructure, technological cooperation, standards, and solutions for countries in Asia, Africa, and Latin America. Formally, these projects are described as support for the digital transformation of developing states, but they deepen partners’ dependence on Chinese technologies and create channels for the spread of AI systems, especially mass-surveillance technologies.
Experts from the Carnegie Endowment, the European Parliament, and Freedom House agree that the main danger of AI in authoritarian regimes is the ability to suppress any protest before it has even properly materialized. AI systems can help authorities detect early signals: unusual online activity, preparations, movement patterns, and links between activists.
All of this has, among other things, already been done in the digital Gulag that Moscow has built. As journalist Andrei Zakharov writes in his new book “Russian Cyberpunk,” Moscow authorities have been collecting comprehensive data on Muscovites since as early as 2014, and have been using AI to bring information from different agencies into a single database. After the 2021 protests, activists were arrested using facial recognition systems. According to Zakharov, in Moscow in 2023, there were already more than 220,000 video devices connected to a unified data storage and processing center.
Now, against the backdrop of sentences of ten-plus years for publishing an article in a foreign journal, for example, the sentences handed down in 2021 to people prosecuted after the so-called “Palace Case” seem small — but this was the first instance in which people in Russia were arrested as a result of video surveillance analysis. Police came to Yevgeny Yesenov’s apartment and arrested him the day after the rally; he became the first prisoner in the case and was sentenced to four years in a penal colony, though he was released on parole after a year and a half. While things like this happened in pre-war Russia, now, as a result of organized surveillance by the “smart city” system, anyone’s address can be found using their face.
After the start of the full-scale invasion of Ukraine, people were arrested in the same way for anti-war graffiti or for setting fire to military enlistment offices — any forms of protest that had been captured and analyzed by the country’s AI surveillance system.
As T-Invariant writes, Russia is also developing AI for drones and cryptography at Moscow State University under the leadership of Putin’s daughter, using Chinese and American gray-market chips and open Western AI technologies. It is also sharing such technologies with China and North Korea.
China and Western Asia
Chinese participation in “smart city” projects in Western Asia is unfolding in a region where, according to the EIU Democracy Index for 2023, 16 out of 20 territories are classified as authoritarian regimes. These projects are most highly concentrated in the Gulf states: the UAE has 11 projects, while Saudi Arabia has seven. Outside this bloc, Turkey plays the most visible role with five projects, while the remaining projects are distributed among Algeria, Egypt, Iraq, Iran, Morocco, and Tunisia.
The latest Chinese concept of the “smart city,” from 2022, involves real-time information monitoring across virtually all aspects of urban life. This technology makes it possible to establish continuous real-time digital monitoring of a city, giving the government the ability to multiply its control over citizens.
China and Russia’s links with Iran
In February 2026, research by the civil society organization Article 19 found that the architecture of internet control in Iran relies heavily on Chinese technologies. The report said that China supplied Iran with facial recognition technologies, internet filtering equipment, and surveillance tools from companies such as Huawei, ZTE, Hikvision, Tiandy, as well as smaller Chinese suppliers.
At the same time, an investigation by Le Monde and Forbidden Stories showed that Iran acquired the Russian facial recognition system FindFace, developed by NtechLab, through Iranian intermediary companies. The software was resold to entities linked to the Iranian state, including the Islamic Revolutionary Guard Corps (IRGC) and the Ministry of Intelligence.
As early as 2019, when Iranian women began receiving fines by mail for not wearing the hijab while in a car, citizens have been talking about their government’s use of digital surveillance. In December 2025, Iran and Russia signed a cooperation agreement in the fields of AI, cybersecurity, the digital economy, blockchain, fintech, and partnerships with the private sector.
Surveillance of people in authoritarian regimes is organized by both cameras and control of internet traffic. The Russian company Protei, headquartered in Jordan but rooted in St. Petersburg, has reportedly supported Iran’s efforts to introduce systems for interception, monitoring, and traffic management, as well as to integrate DPI into network architecture. Its clients include MTN Irancell, Iran’s largest telecom provider, as well as operators and government entities in Tajikistan, Uzbekistan, Pakistan, Kenya, Namibia, Qatar, and other countries.
Taraneh M., an artist living in Iran who was detained during the “Woman, Life, Freedom” protests, said that in 2022 she had prepared for the possibility of arrest by securing her devices, deleting confidential information, and setting self-destruct timers for personal chats. Nevertheless, investigators presented her with a thick dossier of her social media posts, some of which dated back five years. Although she has been released, Taraneh is now afraid to speak on the phone lest she is being monitored.
The China-Russia relationship
Although AI development in China has surpassed Russia’s, cooperation exists in the supply of Chinese technologies, as well as scientific development.
In January 2025, on the heels of Russia’s attempts to create a network of AI alliances with BRICS and other non-Western partners, Vladimir Putin instructed the Russian government and Sberbank to deepen cooperation with China, especially with regard to AI “technological research and development.”
The immediate stimulus was sanctions, Russia’s access to advanced chips and AI equipment was restricted after its invasion of Ukraine, but its dependence on China has persisted in recent years. Supplies for new servers, for example, which Russia does not manufacture, came from China through intermediary firms.
As T-Invariant discovered, however, joint developments are being conducted in two areas: training AI for drones and mass surveillance of people. In November 2025, the creation of two joint laboratories was announced: one for nano-structured optoelectronic materials and devices, and another for AI algorithms and their applications.
North Korea gets involved
Russia and North Korea are cooperating in testing AI in military operations: North Korea supplies weapons that are tested in Russia’s war on Ukraine, and data that may improve North Korean systems, including AI-guided systems, is passed back to North Korea. In May 2026, North Korea tested high-precision cruise missiles and artillery systems with AI guidance.
According to analysts, “AI” likely refers to automatic target recognition and terminal-stage guidance, although the level of technological sophistication remains unclear.
Since late 2023, North Korea has supplied Russia with ballistic missiles and artillery rockets, and their use on the battlefield in Ukraine is believed to have given Pyongyang, and likely Russia, valuable operational data.
The China-Venezuela-Belarus triangle
In Venezuela, authoritarian digital control is being built through international supply chains. Regimes do not necessarily create such systems themselves. Instead, they buy cameras, servers, video analytics platforms, DPI systems, algorithms, and software, often from countries also experiencing digital authoritarianism.
China’s Dahua, despite being added to the U.S. Entity List in 2019 because of its involvement in repression in Xinjiang, retained access to the Venezuelan market and discussed the integration of AI into surveillance systems with the authorities. Possible participation by Huawei, ZTE, and China CAMC in projects is also mentioned, although their role remains unspecified.
Between 2021 and 2023, the Venezuelan interior ministry also held talks with the Belarusian company Synesis, which is under U.S. sanctions for technologies that Belarusian authorities used to identify and arrest protesters. The Kipod software offered to Venezuela included facial recognition; searches for specific people and routes; license plate recognition; detection of weapons, abandoned objects, smoke, and crowds; perimeter control; and monitoring of traffic violations.
Countries on the African continent
A 2026 Institute of Development Studies (IDS) report on surveillance in “smart cities” in Africa showed the spread of Chinese AI-surveillance systems in 11 countries, including Algeria, Egypt, Kenya, Rwanda, Senegal, and Zimbabwe.
Carnegie’s study on the global use of AI in human surveillance predicted that the adoption of such technologies in Africa is likely to grow. Chinese exports of surveillance systems are being used in authoritarian contexts — including in Uganda and Myanmar, often through companies such as Huawei, Hikvision, and Dahua.
As in Western Asia, China-Africa cooperation in AI may look like assistance in developing “smart cities,” but its repressive use is connected with control over citizens.
Central Asia included
Governments in Central Asia have actively purchased Chinese digital public-security systems. A report by the Caspian Policy Center says that in 2013, Tajikistan cooperated with Huawei to create a safe city surveillance system in Dushanbe worth about 22 million USD, largely financed by a Chinese loan.
Other publications note that Chinese companies occupy leading positions in “safe” and “smart city” projects across Eurasia, especially because the region offers relatively open markets for Chinese technologies.
Chinese censorship systems and related tools have also appeared in Kazakhstan, Pakistan, and Ethiopia, but cooperation between China and Central Asia mainly represents the export of AI surveillance technologies under the rhetoric of crime prevention.
What can be done?
Even supposedly democratic states have employed surveillance technologies to spy on their citizens. One of the most functional proposals to hinder the development of such authoritarian digital tools is to make Western technologies open source, and move digital strategies away from sovereignty and towards inclusion of the majority world : not as territories of extraction for the energy and water to run powerful data centers that serving Western AI models, but as actual participants in digital strategies and AI governance.
Such an approach would include sustainable development and practical concerns like agriculture, healthcare, education, transport, urban planning, adaptation to climate change, and many more human-centered priorities.
