AI is neither the autocrat’s silver bullet nor democracy’s death sentence

Artificial intelligence (AI) is widely discussed as a threat to democracy—through deepfakes, disinformation, and the erosion of public trust.

Less examined, but often implicitly assumed, is the corollary: that AI asymmetrically benefits autocracies, which can deploy these tools without the constraints of institutional oversight or public accountability. According to this view, AI supercharges state control while undermining the open societies that might resist it.

The reality is more complicated: AI does not map neatly onto a spectrum running from democratic vulnerability to authoritarian advantage. It introduces pressures, disruptions, and opportunities across regime types, and its political effects depend heavily on context—existing institutional capacity, the strength of civil society, and the specific ways the technology is adopted. Democracies face genuine risks, but autocracies are not simply collecting a dividend.

Consider China, which offers perhaps the strongest test case for the thesis that AI empowers authoritarian control. Over the past decade, Beijing has assembled one of the most sophisticated digital censorship and surveillance apparatuses in history: facial recognition cameras blanket public spaces, AI-powered content filters scrub social media in real time, and a social credit system tracks behaviour across domains. If any state should have cracked the code of algorithmic control, it is the People’s Republic of China (PRC). And yet the evidence is ambiguous. According to the China Dissent Monitor, incidents of public dissent have continued to rise even as digital authoritarianism has deepened. The Chinese Communist Party (CCP) itself remains deeply concerned about popular mobilization; internal directives consistently flag “social stability” as a top governance concern. These seemingly contradictory trends raise questions about how effective new technology really is as a tool of political control. The world’s most digitally surveilled population has not stopped taking to the streets, and while surveillance may well be suppressing some dissent that would otherwise occur, the gap between the technology’s promise and its observable results is difficult to ignore.

Where digital authoritarianism is working

In this piece, when we discuss AI-enabled authoritarianism, we do not mean AI in isolation. Artificial intelligence is one layer in a broader stack of digital control technologies—e.g., biometric databases, internet filtering, device tracking, integrated payment systems—many of which predate recent advances in machine learning. What AI adds is speed, scale, and adaptive capacity: the ability to process vast data streams in real time, to automate decisions that once required human judgment, and to refine control mechanisms through continuous feedback. The political effects we describe in this piece arise from this fuller technological ecosystem.

Certain dimensions of technology-enabled control are functioning exactly as intended. Surveillance infrastructure—facial recognition, movement tracking, predictive policing algorithms—has made it significantly harder for dissidents to organize anonymously. Most notably in Xinjiang, a comprehensive digital monitoring system has been deployed against the Uyghur population, integrating biometric data, phone tracking, and AI-driven risk scoring into an architecture of repression that operates at a scale previously unimaginable.

Information control has also become more sophisticated. China’s Great Firewall no longer merely blocks foreign content; it actively shapes the domestic information environment through algorithmic curation, automated censorship, and the strategic deployment of state-aligned narratives. The result is not just the suppression of dissent but the erosion of shared foundations; citizens increasingly struggle to distinguish reliable information from state propaganda, which may be precisely the point.

These digital tools are, moreover, embedded in physical infrastructure. Smart city projects, transportation networks, and payment systems all generate data streams that feed back into the surveillance apparatus. The line between public service provision and population monitoring has become vanishingly thin. The CCP’s capacity for social control further rests on a dense network of human intermediaries, such as neighbourhood committees, local cadres, community-level police stations, and surveillance operatives, that have long penetrated the fabric of everyday life.

The limits of digital control

Yet controlling the digital world has proven insufficient for fully controlling offline actions; even in China, one of the world’s most digitally connected societies. Protests continue to erupt over issues including land seizures, wage theft, environmental degradation, and local corruption. The anti-lockdown protest movement of 2022, triggered by mounting frustration with zero-COVID policies, grew over months across the country despite extensive online censorship and surveillance. The movement ultimately culminated in the “white paper” protests, characterized by direct criticism of the central leadership and the political system. This significant decentralized movement emerged despite heavy online censorship.

Why? Part of the answer is that grievances rooted in material conditions—unpaid wages, land grabs, demolished homes, polluted water—do not disappear simply because algorithms suppress their online expression. Surveillance can raise the cost of organising, but it cannot eliminate the underlying motivations. When discontent runs deep enough, people find ways to act: through real-world social groups, through word of mouth, or through deliberate evasion of digital controls.

There is also an irony embedded in the technology itself. The same digital platforms that the state uses for surveillance and propaganda also serve as spaces where citizens observe each other’s grievances, however briefly, before censors intervene. A post can become viral within hours, and the act of censoring it can itself become a signal that something has occurred that is worth paying attention to. There is a further, less intuitive dynamic at work. Stronger technological tools of control may themselves contribute to declining public trust. When citizens know they are being surveilled, when they encounter automated censorship in their daily interactions with digital platforms, and when they suspect that the information environment is being manipulated, their confidence in institutions—including the institutions doing the surveilling—can erode. Research has shown an inverse relationship between public trust and protest participation, which suggests that the very technologies the CCP deploys to maintain stability may, over time, be undermining one of its preconditions. The tools of control, in other words, may be generating the conditions for the discontent they are designed to suppress.

More broadly, both democracies and autocracies depend on forms of public trust, even if the nature of that trust differs across regime types. Democracies rely on trust in institutions and information, while autocracies often depend on trust in ideology, party structures, or the credibility of repression itself. If AI contributes to a broader erosion of trust through misinformation and uncertainty, this may create governance challenges not only for democracies, but for autocracies as well.

Exporting the model across Asia (and beyond)

China’s digital authoritarian toolkit is not staying within its borders. Through the Belt and Road Initiative, technology partnerships, and direct sales of surveillance equipment, Beijing is exporting the infrastructure and the governance logic of its model to countries across Asia and beyond.

The appeal is broad precisely because the importing countries span a wide spectrum of regime types. Authoritarian states like Myanmar’s military junta have an obvious interest in Chinese-style population monitoring. But semi-democratic and democratic governments—Sri Lanka, the Philippines, Indonesia—are also interested in acquiring facial recognition systems, safe city platforms, and AI-driven content moderation tools. The common thread is arguably not ideology but a shared desire among governing elites for greater control over their populations, whether framed as national security, public order, or modernisation.

It is in these countries that AI may pose the greatest risk of accelerating authoritarian trends. China built its surveillance apparatus over decades, layering digital tools on top of an already extensive infrastructure of human intermediaries and institutional control. Countries that lack this pre-existing architecture have historically faced high barriers to comprehensive population monitoring. AI-enabled surveillance packages, sold as turnkey systems and bundled with safe city rhetoric, dramatically lower those barriers. A government that could not have built China’s system from scratch can now purchase key components of it off the shelf. The result is that AI may not give already consolidated autocracies the silver bullet they are looking for, while simultaneously offering aspiring ones a shortcut they would not otherwise have had.

This diffusion matters because it embeds authoritarian governance assumptions into technical systems that may outlast any particular government. Once a facial recognition network is installed and integrated with national ID databases, its removal becomes a political and logistical challenge regardless of whether the regime that purchased it remains in power.

Getting our priorities right

This analysis has direct implications for how democracies and civil society should allocate their attention and resources in responding to digital authoritarianism.

The current discourse is heavily weighted toward the threats AI poses to democratic societies—and those threats are genuine. But it is also shaped by a parallel assumption: that AI is making autocracies decisively stronger, more efficient, and more durable. Our analysis suggests that neither framing captures the full picture. AI has not given consolidated autocracies the totalizing control that some feared, but nor has it weakened them — it has changed the terrain on which both control and resistance operate, in ways that remain contested and context-dependent. An exclusive focus on defending democracy at home risks neglecting the arena where the stakes may be highest: the global contest over whether authoritarian governance models, enabled and exported by AI-powered surveillance, become the default infrastructure of state power across the developing world.

We suggest several policy priorities to address these challenges.

First, democratic governments and international organisations need a much clearer framework for where tools such as facial recognition software and AI-driven surveillance are permissible and where they are not. The absence of such guardrails leaves a vacuum that authoritarian exporters are filling with their own norms.

Second, the crisis of public trust demands attention as an issue in its own right, not merely as a side effect of AI. Rebuilding epistemic infrastructure, including reliable media, transparent institutions, and information literacy, is harder and slower than deploying content filters, but it is the only durable response to a world in which generative AI makes deception cheap.

Third, the role of private technology companies cannot be ignored. The same firms that sell surveillance systems to authoritarian and semi-authoritarian governments also operate in democratic markets and shape global technology standards. Where these companies stand, or can be pressed to stand, on questions of human rights, data protection, and civil liberties is a matter of geopolitical consequence, not just corporate social responsibility.

Fourth, democratic governments and international civil society must not, on the assumption that AI makes autocracies omnipotent, give up on supporting the people in those countries who continue to speak out on public issues and demand accountability despite their governments. The Chinese government suppressing the visibility of dissent does not indicate Chinese citizens have greater trust in the government or have abandoned bottom-up participation.

Conclusion

The battle against digital authoritarianism will not be won by focusing on the wrong front. China’s experience—where protest persists and even rises despite one of the world’s most advanced surveillance systems—is a reminder that technological power and political control are not the same thing.

But the more pressing concern may lie elsewhere. While AI has not perfected authoritarian control where it already exists, it is lowering the barriers to authoritarian control where it does not yet exist. For semi-democratic and hybrid regimes across Asia and beyond, AI-enabled surveillance tools offer a shortcut to capabilities that would previously have taken decades and vast institutional infrastructure to develop. The evidence suggests that for countries in the process of autocratizing, AI is on balance more enabling than constraining—and it is in these contexts that the technology's political consequences may prove most consequential.

What is at stake is whether the international community recognizes this complexity in time to act on it. If democratic societies spend their energy exclusively on domestic AI regulation while authoritarian surveillance infrastructure quietly becomes the backbone of governance across the developing world, the most consequential front in the struggle for political freedom will have been conceded without a fight. Our three-year research project at PRIO intends to confront these questions directly, investigating how democracy and civil society actors can best counter digital authoritarianism — not by assuming that AI is uniformly catastrophic, but by understanding precisely where, how, and for whom it changes the balance of power.

• The authors are all members of the AsiaTech project at PRIO