Two weeks after Moscow launched its full-scale invasion of Ukraine in March 2022, a dramatic breach pierced the heavily guarded facade of Russian state television. Marina Ovsyannikova, an editor at Channel One, burst live onto the set of the evening newscast during a broadcast. Positioned directly behind the news anchor’s head, she held up a hand-drawn sign bearing a clear, defiant message: “Stop the war. Don’t believe propaganda. They’re lying to you!” As cameras scrambled to redirect their angles, she shouted, “No to war!” repeatedly before being physically dragged away by security personnel.

That iconic moment of dissent remains a rare anomaly. No one has dared to openly protest the war on Russian television in the years since, an absence of defiance largely secured by tightened internal security, severe legislative penalties, and a pervasive climate of fear. Yet, by October 2024, Margarita Simonyan—one of Russia’s most prominent and influential chief propagandists—offered a radically different explanation for the quiet screens. During a candid television interview, she revealed that a growing number of anchors across the state-backed RT network are not real human beings at all.

“That face doesn’t exist,” Simonyan explained to her interviewer, noting that both the visual representation and the spoken word are entirely synthetic. “We generated the voice and everything else.”

These anchors, she confirmed, are produced entirely by artificial intelligence. In a subsequent interview with the Russian business newspaper Vedomosti, Simonyan laid out the administrative logic behind the technological pivot with chilling frankness. “These anchors don’t need a salary or insurance,” she remarked. “They won’t get arrested, and the police won’t search their homes. It’s all wonderful and terrifying at the same time.”

This dual sentiment captures the core promise that artificial intelligence extends to every autocratic ruler across the globe today: the seductive ability to maintain absolute political control without having to delegate vital tasks to unreliable, unpredictable human subordinates. It is an exceedingly attractive prospect for authoritarian governance. Regardless of how secure or powerful a dictator might be, authoritarian leaders have historically depended on a vast infrastructure of human agents—censors, propagandists, security guards, police officers, intelligence analysts, and local bureaucrats—to execute their commands and enforce state policy.

Yet, human subordinates inherently pose a permanent, structural threat to any regime. They possess the capacity to shirk duties, misappropriate funds, leak sensitive state secrets, lie to their superiors, or quietly conspire against the boss. Consequently, one of the oldest and most persistent dilemmas facing autocrats throughout history has been figuring out how to empower agents sufficiently to carry out state orders without simultaneously enabling those same agents to turn against the supreme leader.

Autocrats have traditionally managed this delicate trade-off by populating key government positions with career mediocrities whose personal and professional limitations serve as, in the words of political theorist Hannah Arendt, “the best guarantee of their loyalty.” However, this systemic reliance on incompetence inevitably makes it significantly harder for dictatorships to govern efficiently, manage complex economies, and survive severe political or social crises.

Artificial intelligence now appears to offer modern autocrats two distinct pathways to resolve this long-standing governance dilemma. First, it grants leaders the capacity to continuously monitor, evaluate, and discipline their followers in real time. Subtle shifts in a local official’s spending habits, lifestyle changes, or unexplained modifications to meeting schedules can now be flagged automatically by sophisticated AI surveillance systems, entirely bypassing the need for self-interested human informants.

Second, AI promises to eliminate underlings altogether by automating complex administrative tasks that once required human discretion and personal judgment. By deploying faithful algorithms that cannot be bribed, coerced, or manipulated, rulers can replace unpredictable intermediaries with automated systems. Officials within autocratic regimes have increasingly voiced this ambition openly. Maksut Shadaev, Russia’s Digital Development Minister, told attendees at a Moscow data forum in April 2025 that artificial intelligence “can definitely replace half of officials,” before adding, “maybe slightly more.”

Yet, these sweeping promises of a self-running, frictionless state are ultimately illusory, because advanced artificial intelligence systems are never truly autonomous. Complex machine-learning systems must be built, calibrated, and continuously maintained by an elite cadre of software engineers and data scientists whose specialized technical expertise political leaders are fundamentally incapable of evaluating.

The ruler who turns to artificial intelligence specifically to reduce their dependence on unreliable human subordinates may ultimately find themselves relying, far more blindly than before, on the tiny fraction of technical specialists who keep the digital infrastructure operational. This newly empowered digital Praetorian Guard carries many of the exact same systemic weaknesses as the historical guards that preceded them, quietly transforming into a potent institutional threat to the ruler.

EYES AND EARS

When it comes to developing and deploying artificial intelligence systems capable of monitoring or replacing human subordinates, China currently leads the global field. Advanced smart-city programs operating in Beijing, Shanghai, and other major urban centers utilize AI algorithms to track administrative performance and centralize the oversight of local officials who once operated in comfortable obscurity. This technological shift subjects local bureaucrats to unyielding algorithmic performance reviews.

Over the past decade, China also developed and deployed what it termed a “Zero Trust AI system” across approximately 30 counties and cities. This aggressive platform cross-referenced data from more than 150 separate government databases with the explicit goal of detecting embezzlement, corruption, and nepotism among local officials. The system arguably worked too well, flagging roughly 8,700 officials before local governments, facing intense political pressure from the very bureaucrats the technology was monitoring, began scaling the program back.

Despite those local rollbacks, Beijing has aggressively maintained its modern e-government portals. These platforms utilize automated systems for issuing official documents, licenses, and permits, effectively minimizing the face-to-face interactions that historically bred petty corruption and personal favoritism. Meanwhile, the Chinese Communist Party has plowed ahead with extensive research into what state officials describe as “thought management”—exploring how to leverage artificial intelligence to craft microtargeted propaganda. China’s vast online army of internet commentators, which once relied entirely on paid human workers, is already being systematically replaced by automated bots.

Russia has pursued these same technological objectives with a remarkably similar fervor. Moscow’s expansive citywide facial recognition network, integrated across a staggering 200,000 public cameras, was deployed extensively to identify and apprehend protesters during the anti-government demonstrations of 2021. The central government has steadily incorporated automated data collection and aggregation protocols designed to bypass regional officials who might otherwise taint administrative data or manipulate it to promote their own personal political interests.

In 2023, Russia’s federal internet regulator launched Oculus, an advanced AI system engineered to scan hundreds of thousands of digital images and texts per day for prohibited content, including unauthorized political material. This automated scanning operates at an order of magnitude far beyond the processing capabilities of traditional human censors. Similarly, the Russian security service’s internal tool, known as Meliorator, has been utilized to automatically generate more than a thousand profiles of fictitious American citizens—predominantly on the social media platform X, formerly known as Twitter—to amplify official Kremlin narratives at a fraction of the financial and logistical cost required by traditional human-operated troll farms.

Other autocratic nations are rapidly following in the footsteps carved by Beijing and Moscow. In April 2025, the United Arab Emirates established a specialized Regulatory Intelligence Office that leverages artificial intelligence to draft, review, and amend federal laws and regulations—responsibilities historically managed by legislative staff and legal drafters. Just a year later, UAE Prime Minister Sheikh Mohammed bin Rashid al-Maktoum announced a sweeping national initiative stating that autonomous AI agents would take over half of all federal government operations within a span of two years. He explicitly emphasized that “the performance of ministers, directors general, and entities will be assessed based on their ability to adopt this transformation.”

These technological innovations may successfully reduce the sheer number of traditional bureaucrats that autocrats need to hire and manage. However, they cannot actually materialize what dictators desire most: a completely self-running state devoid of human friction. Even AI-generated television anchors still require specialized human technicians to build, render, and maintain them. Likewise, automated censorship systems require human engineers to retrain algorithms as the linguistic vocabulary of political dissent inevitably shifts and evolves.

Behind every sophisticated AI model lies a small, insular group of software engineers and data scientists performing essential maintenance work. Their labor is so deeply technical that supreme political leaders cannot comprehend its inner workings.

This profound technical dependence inevitably morphs into a formidable new form of political power. The engineers who maintain an autocrat’s surveillance and censorship apparatus, for instance, retain the unique ability to quietly shape precisely what information the ruler is allowed to see. They decide which incoming data streams are important enough to pass up the chain of command and in what visual or analytical form those reports should be presented.

These specialists determine which potential political threats get flagged by the system, and they retain the power to subtly adjust the underlying algorithms that select which digital content gets suppressed and which political dissidents get arrested. They can execute these alterations without anyone inside the palace noticing the manipulation. Autocrats who turn to artificial intelligence to escape their historical dependence on unreliable human subordinates have merely transferred their vulnerabilities onto an entirely new, highly specialized group of technical officials.

In practice, this new digital Praetorian Guard could prove significantly more dangerous to a regime than previous traditional political elites. This heightened risk stems partly from the inherent opacity of complex AI technology, but also from the remarkably small size of this technical clique. While ancient Roman emperors typically relied on tens of thousands of soldiers serving within the physical Praetorian Guard, and modern autocrats rely on vast standing armies, managing an advanced artificial intelligence infrastructure requires only a tiny, concentrated clique of skilled engineers.

While this concentration might initially appear beneficial to a ruler because there are fewer individuals to monitor, it simultaneously creates a dangerous single point of failure. Disrupting or subverting the state apparatus no longer requires a mass mobilization; it requires only a handful of cooperative engineers, while also making it vastly easier for those members to covertly scheme against supreme leadership. Furthermore, members of this new digital guard are virtually irreplaceable on short notice. While an autocrat can readily replace military generals without destroying the broader army, operating complex, proprietary machine-learning systems requires such a deeply specialized set of cognitive skills that engineers are exceptionally difficult to swap out without causing catastrophic institutional disruptions, thereby granting these technical actors immense political leverage.

THE INDISPENSABLES

The critical importance of technology workers to modern autocracies has already manifested in stark ways. When skilled IT specialists began fleeing Russia in large waves following the full-scale invasion of Ukraine in February 2022, the Kremlin reacted not with immediate threats of imprisonment or violence, but with sweeping economic inducements. The government offered tech workers official deferments from military conscription, exempted their corporate firms from corporate taxes, and provided heavily subsidized mortgages.

When Russian President Vladimir Putin subsequently ordered the mobilization of 300,000 reservists seven months later, domestic IT workers were granted comprehensive, blanket waivers from the draft. While this level of administrative protection might initially seem disproportionate given that an established AI system requires only a relative handful of engineers to operate, a fragile regime cannot know in advance precisely which individual engineers it will ultimately need during a crisis. Because it cannot train replacements quickly, the state is forced to protect and retain the entire broader talent pool from which those few indispensable specialists are drawn.

Despite these aggressive economic measures, the financial incentives failed to stem the outward migration of roughly 100,000 technology specialists—amounting to approximately one-tenth of Russia’s entire domestic tech workforce—over the course of 2022. Yet, Moscow steadfastly stuck to bribery and institutional protection because technical expertise is extremely difficult to conscript successfully at gunpoint, and because forced experts might be far less likely to follow sensitive instructions faithfully.

Even so, financial compensation and institutional privilege cannot guarantee that these technical elites will remain politically loyal. Autocrats would be wise to recall the historical lesson taught by the original Praetorian Guard of ancient Rome. For a period, the guard provided Roman emperors with reliable protection and personal security. But eventually, the praetorians discovered the true value of their own indispensability, and by the second century, they were openly auctioning off the imperial throne to the highest financial bidder.

This new class of digital elites possesses the capability to do precisely the same thing. In June 2023, when a heavily armed column of Yevgeny Prigozhin’s Wagner paramilitary company rolled up the highway toward Moscow in an armed mutiny, President Putin relied heavily on his security services to inform him of developments on the ground. Future autocratic rulers occupying Putin’s position will no longer receive such urgent warnings primarily through human channels; instead, those warnings will arrive mediated through machines—intercepted communications sorted by complex software, live camera feeds filtered through automated recognition systems, and regional intelligence reports compiled automatically into executive dashboards.

The engineers who run those underlying digital systems could easily strike a clandestine deal with an upstart political challenger and subsequently begin dragging their feet. They could intentionally delay incoming data feeds, let critical security alerts arrive a few hours late, degrade live camera feeds at inconvenient tactical moments, or cause a facial recognition system to suddenly stop functioning altogether. In such a scenario, the ruler would be operating completely blind. The ancient Praetorian Guard did not actually need to assassinate Emperor Nero to replace him; they simply had to abandon him and shift their allegiance to a rival claimant.

While these programmers are unlikely to seize direct political power for themselves because they generally lack the instruments that coup leaders ultimately require—namely, physical weapons—there is already concrete historical precedent for software engineers using their technical leverage to shape major leadership challenges. In August 1991, when hardline elements of the Soviet military attempted to stage a coup to seize control from Soviet leader Mikhail Gorbachev, a small group of defiant programmers operating the Relcom computer network successfully kept independent information flowing abroad while relaying Russian President Boris Yeltsin’s decrees to audiences both inside the Soviet Union and across the international community. The coup plotters found themselves entirely unable to stop the news of Yeltsin’s public defiance from spreading digitally, and the authoritarian putsch collapsed within three days.

SLIMMING DOWN

Artificial intelligence will never realistically allow autocrats to dismiss all of their human enforcers. Someone must still physically carry out arrests, guard borders, and run physical prisons, and autocrats will continue to buy baseline political loyalty by offering supporters lucrative state jobs. However, automation will successfully allow authoritarian regimes to significantly downsize their administrative bloat.

Ultimately, the absolute number of state officials matters far less to a dictator than the efficiency with which the ruler can oversee them. Keeping track of scheming, incompetent, or disloyal subordinates has historically remained the single greatest administrative burden weighing upon any autocrat. Artificial intelligence promises to lighten that heavy load, enabling rulers to monitor their officials far more closely and continuously without having to invest as much direct, active personal attention.

In the near term, therefore, artificial intelligence may deliver substantial benefits to autocratic regimes. Despite the numerous logistical and technical obstacles involved in deployment, the technology is already providing authoritarian governments with tangible operational advantages, including cheaper mass surveillance, smarter and more pervasive censorship, and far fewer human subordinates to fear and distrust.

Yet, for autocratic rulers, the deep temptations of artificial intelligence may ultimately re-create the very same structural trap they are actively trying to escape. The more an authoritarian regime depends on artificial intelligence to sustain its grip on power, the more deeply it becomes dependent on the specialized people who keep that artificial intelligence running. And those technical specialists, much like every other praetorian class recorded in human history, will quickly discover precisely what their indispensable expertise is truly worth.