Two weeks after Moscow’s full-scale invasion of Ukraine in March 2022, a dramatic breach pierced the tightly controlled facade of Russian state television. Marina Ovsyannikova, an editor for Channel One, burst onto the set during the live evening newscast. Positioned directly behind the news anchor’s head, she held aloft a hand-drawn sign bearing a stark message: “Stop the war. Don’t believe propaganda. They’re lying to you!” She shouted “No to war!” repeatedly before security personnel rushed forward and dragged her off the set.
That act of defiance remains a stark anomaly in the modern history of Russian state broadcasting. No one has dared to protest the war on Russian television since that night, a silence enforced by tightened security protocols, severe legislation, and a pervasive climate of fear. Yet, by October 2024, Margarita Simonyan, one of the Kremlin’s chief propagandists and the head of RT, offered an entirely different, technologically driven explanation for the absence of dissent on the network.
During an interview, Simonyan revealed that a growing number of RT’s news anchors were no longer human beings. “That face doesn’t exist,” she explained to her interviewer. “We generated the voice and everything else.” These anchors, she acknowledged, were produced entirely by artificial intelligence. Elaborating on this development in a follow-up interview with the business newspaper Vedomosti, Simonyan spelled out the underlying logic of integrating synthetic media into state-controlled broadcasting. “These anchors don’t need a salary or insurance,” she noted. “They won’t get arrested, and the police won’t search their homes. It’s all wonderful and terrifying at the same time.”
This dynamic captures the core promise that artificial intelligence extends to every autocratic ruler across the globe: the tantalizing prospect of maintaining absolute political control without the persistent hazard of delegating critical tasks to unreliable human subordinates. It is an extremely attractive vision of governance. Regardless of their power, dictators have always depended on a vast network of human subordinates—including censors, propagandists, security guards, police officers, intelligence analysts, and local bureaucrats—to execute their orders and enforce state policy.
Yet, human subordinates inevitably introduce structural vulnerabilities. They can shirk their duties, steal state resources, lie to their superiors, leak sensitive information, or conspire behind closed doors against the regime. Consequently, one of the oldest dilemmas facing authoritarian leaders has centered on how to empower agents to carry out directives without simultaneously enabling those same agents to turn against the political leadership.
Historically, autocrats have managed this delicate tradeoff by staffing critical government positions with mediocrities whose professional limitations ensure their dependence. As the political theorist Hannah Arendt famously observed, this institutionalized incompetence serves as “the best guarantee of their loyalty.” However, this reliance on mediocre administrators inevitably degrades the overall administrative capacity of the state, making it far more difficult for dictatorships to govern effectively or survive acute domestic crises.
Artificial intelligence appears to offer modern autocrats a way out of this long-standing governance dilemma through two distinct mechanisms. First, machine-learning systems provide rulers with the unprecedented ability to monitor, evaluate, and discipline their followers in real time. A sudden shift in a local official’s spending patterns, travel habits, or meeting schedules can be flagged automatically by an algorithmic system, completely bypassing the need for self-interested human informants or whistleblowers.
Second, AI promises to eliminate underlings altogether by automating complex tasks that once required human discretion. By deploying algorithms, rulers can replace unpredictable intermediaries with faithful digital tools that cannot be bribed, coerced, or manipulated. Officials within autocratic governments have openly voiced this ambition. Reflecting on the trajectory of state modernization, Russia’s Digital Development Minister, Maksut Shadaev, told an audience at a Moscow data forum in April 2025 that artificial intelligence “can definitely replace half of officials,” before adding, “maybe slightly more.”
Yet, these visions of a self-running, frictionless state are fundamentally illusory, because artificial intelligence systems are never truly autonomous. Underlying algorithms must be meticulously built, calibrated, and continuously maintained by a specialized cadre of engineers, programmers, and data scientists whose technical expertise political leaders are utterly incapable of evaluating. The autocratic ruler who turns to artificial intelligence to reduce their dependence on unreliable human beings may ultimately find themselves relying, far more blindly than before, on the tiny group of technical specialists who keep the digital infrastructure operational. This newly empowered digital elite may well possess the exact same systemic weaknesses as the traditional political enforcers they replace, ultimately reconstituting the very threat the ruler sought to escape.
EYES AND EARS
China currently leads the global field in developing and deploying advanced artificial intelligence systems designed to monitor, evaluate, and replace human subordinates within the administrative apparatus. Advanced smart-city programs operating in Beijing, Shanghai, and other major urban centers utilize AI to track performance metrics and centralize administrative oversight over local officials. These officials, who once operated within comfortable obscurity, are now routinely subjected to rigorous algorithmic performance reviews.
Furthermore, China developed an experimental framework known as the “Zero Trust AI system,” which was deployed experimentally across approximately 30 counties and cities over the past decade. The system was designed to cross-reference more than 150 disparate government databases simultaneously to detect patterns of embezzlement, corruption, and nepotism among local bureaucrats. Ultimately, the system proved effective to a fault, automatically flagging some 8,700 local officials before regional governments—facing intense institutional pushback from the very bureaucrats the technology was monitoring—began scaling the program back.
Despite such pushbacks, Beijing has maintained its commitment to e-government portals that utilize automated processing systems for issuing official documents and administrative permits, thereby reducing the face-to-face interactions that historically bred petty corruption and personal favoritism. Simultaneously, the Chinese Communist Party has heavily invested in researching what state directives term “thought management”—exploring how AI can generate microtargeted propaganda tailored to specific demographics. China’s sprawling apparatus of internet commentators, which once relied entirely on paid human workers, is increasingly being transitioned to automated bot networks.
Russia has pursued comparable strategic objectives with similar institutional fervor. Moscow’s citywide facial recognition infrastructure, integrated across a network of roughly 200,000 public cameras, was extensively utilized to identify, track, and apprehend participants during the 2021 anti-government demonstrations. Additionally, the federal government has increasingly relied on automated data collection and aggregation systems to bypass regional officials who might otherwise distort reported metrics or exploit local data to advance their own political standing.
In 2023, Russia’s federal internet regulatory agency launched Oculus, an advanced AI system capable of scanning hundreds of thousands of digital images daily for prohibited content, including unauthorized political commentary. This automated capacity operates at an order of magnitude far exceeding what human censors could ever hope to process. Similarly, the Russian security services developed a tool called Meliorator, which has been deployed to generate over a thousand distinct profiles of fictitious American citizens—primarily on the social media platform X, formerly known as Twitter—to disseminate Kremlin narratives at a fraction of the financial cost required to sustain traditional human troll farms.
Other authoritarian states are rapidly following the precedents established by Beijing and Moscow. In April 2025, the United Arab Emirates established a dedicated Regulatory Intelligence Office that leverages artificial intelligence to draft and amend federal legislation—responsibilities historically managed by specialized legislative staff. One year later, UAE Prime Minister Sheikh Mohammed bin Rashid al-Maktoum announced that autonomous AI agents would assume direct control over half of all federal government operations within a two-year timeframe. He explicitly stated that “the performance of ministers, directors general, and entities will be assessed based on their ability to adopt this transformation.”
These systemic innovations may successfully reduce the sheer headcount of administrative bureaucrats that autocratic leaders must maintain on the state payroll. However, they cannot manufacture what dictators desire above all else: a truly self-running administrative state. Even synthetic, AI-generated television anchors require human technicians to construct and maintain their underlying rendering models, and automated censorship engines require human experts to retrain them as the vernacular of political dissent continually shifts. Behind every sophisticated artificial intelligence model stands a small, indispensable group of engineers and data scientists performing critical maintenance work whose technical depth remains entirely opaque to political leaders.
This technical dependence invariably transforms into a potent form of political leverage. The engineers who maintain an autocrat’s digital surveillance and filtering apparatus possess the capacity to shape precisely what the ruler is allowed to see. They determine which incoming information streams are deemed important enough to escalate to the leadership and dictate the format in which those reports are presented. They configure the algorithmic thresholds that determine which security threats are flagged, and they possess the technical latitude to quietly adjust the parameters governing what content is suppressed and who gets arrested. Crucially, they can execute these adjustments without anyone inside the halls of power noticing. Autocrats who turn to artificial intelligence to escape their historical dependence on unpredictable human subordinates have merely transferred their fundamental vulnerabilities onto a new, highly specialized class of operators.
In practical terms, this emerging digital elite could prove significantly more dangerous to a regime than previous ruling cliques. This vulnerability stems partly from the inherent opacity of complex AI technologies, but also from the remarkably small size of this technological caste. While Roman emperors typically maintained tens of thousands of soldiers within the Praetorian Guard, and modern autocrats rely on expansive standing security forces, the governance of advanced machine-learning infrastructure requires only a tight-knit clique of highly skilled engineers.
While this concentration may initially appeal to rulers by minimizing the number of individuals they need to monitor, it simultaneously concentrates catastrophic points of failure. Disrupting or subverting the system no longer requires a broad conspiracy; it requires only a handful of compromised or disgruntled engineers, while making it far easier for members of this technical elite to coordinate their actions against the supreme leader.
Furthermore, members of this new digital guard are virtually irreplaceable during a crisis. While an autocrat can readily replace military generals without fundamentally destroying the operational capacity of the armed forces, administering and repairing complex machine-learning architectures demands a hyper-specialized skillset. Because such expertise cannot be rapidly substituted without catastrophic system disruptions, these engineers command formidable leverage over the regime.
THE INDISPENSABLES
The critical importance of technical workers to the survival of autocratic governance has already manifested in stark ways. When thousands of IT specialists fled Russia following the initiation of the full-scale invasion of Ukraine in February 2022, the Kremlin’s response avoided heavy-handed coercion, opting instead for a campaign of state inducements. The government offered domestic tech workers official deferments from military conscription, exempted their employing firms from corporate taxes, and heavily subsidized residential mortgages.
When Russian President Vladimir Putin subsequently ordered the mobilization of 300,000 reservists seven months later, technology sector workers were granted comprehensive, blanket waivers from active deployment. While this protective stance might appear excessive given that an automated AI infrastructure theoretically requires only a small fraction of engineers to maintain, an authoritarian regime cannot easily predict in advance which specific specialists it will require, nor can it train qualified replacements on short notice. Consequently, the state is forced to protect and retain the entire domestic talent pool from which those critical few are ultimately drawn.
Despite these extensive government incentives, these measures failed to prevent the departure of approximately 100,000 IT specialists—representing roughly one-tenth of Russia’s entire technology workforce—over the course of 2022. Yet, Moscow steadfastly maintained its reliance on financial bribery and institutional privilege, recognizing that technical expertise cannot be effectively conscripted at gunpoint, and that forced experts would be far less likely to execute directives faithfully.
Even so, generous financial compensation and social privilege cannot guarantee the long-term political loyalty of these technical elites. Autocratic leaders would do well to recall the historical trajectory of the original Roman Praetorian Guard. Initially established to provide Roman emperors with absolute protection and security, the praetorians eventually recognized the extent of their own indispensability. By the second century, they were openly auctioning off the imperial throne to the highest bidder.
A modern technical elite possesses the capacity to enact a similar betrayal. During the armed mutiny in June 2023, when a column of fighters belonging to Yevgeny Prigozhin’s Wagner paramilitary company advanced rapidly up the highway toward Moscow, President Putin relied heavily on his intelligence and security services to provide real-time situational awareness. Future autocratic rulers facing similar insurrections will receive those warnings primarily through automated systems: intercepted communications sorted by natural language software, live surveillance camera feeds filtered through facial recognition algorithms, and regional intelligence reports automatically compiled into executive dashboards.
The software engineers who maintain those critical systems could easily strike a clandestine deal with an upstart political challenger and selectively degrade system performance. They could deliberately delay incoming data feeds, allow security alerts to arrive hours late, introduce deliberate distortions into surveillance streams at strategically inconvenient moments, or cause facial recognition networks to experience sudden, unexplained outages. Under such a scenario, the ruler would be rendered effectively blind. The ancient Praetorian Guard did not need to physically assassinate Emperor Nero to replace him; they simply had to abandon him in his hour of need and shift their allegiance to a rival claimant.
While these computer programmers are unlikely to seize direct political power for themselves—largely because they lack the traditional instruments of physical coercion typically required by coup leaders—historical precedent already demonstrates the capacity of technical specialists to shape high-stakes leadership crises. In August 1991, when hardline elements within the Soviet military attempted to execute a coup d’état against Soviet leader Mikhail Gorbachev, a small group of computer programmers operating the Relcom network successfully kept independent information flowing abroad while simultaneously relaying Russian President Boris Yeltsin’s decrees to audiences both inside the country and across the international community. Unable to suppress the rapid dissemination of news detailing Yeltsin’s public defiance, the plotters watched their operational control disintegrate, causing the coup to collapse within three days.
SLIMMING DOWN
Artificial intelligence will not enable autocrats to completely dismantle their traditional apparatus of state enforcers. Human personnel remain an absolute necessity to conduct physical arrests, manage sprawling prison complexes, and secure ongoing political support by distributing patronage appointments throughout the state apparatus.
Nevertheless, technological automation will allow authoritarian regimes to downsize their administrative footprint. Ultimately, the absolute number of state officials matters far less to a dictator than the efficiency and reliability with which the ruler can oversee them. Maintaining continuous surveillance over scheming, corrupt, or incompetent subordinates has historically constituted the single greatest operational burden for any autocrat. Artificial intelligence promises to lighten this heavy burden, enabling rulers to monitor their administrative officials much more closely without needing to invest direct, continuous personal attention.
In the near term, therefore, artificial intelligence may yield substantial political benefits for autocratic regimes. Despite the numerous logistical and technical obstacles involved in large-scale deployment, the technology is already delivering tangible systemic advantages: cheaper and more pervasive surveillance, increasingly sophisticated automated censorship, and a significantly reduced reliance on unpredictable human subordinates who must be constantly managed and mistrusted.
Yet, for authoritarian rulers, the siren song of artificial intelligence threatens to recreate the exact institutional trap they are actively seeking to escape. The more deeply a regime binds its survival to artificial intelligence, the more profoundly it becomes dependent on the small group of specialized human beings who keep that technology running. And those technical specialists, much like every other praetorian class recorded in political history, will soon discover the true value of their indispensability.
