Authoritarianism, Now With Better Customer Service

One often overlooked but important political implication of AI for authoritarian politics may not be spectacular repression. It may be a subtler shift, the normalization of control through administration. AI allows regimes to present surveillance as efficiency, policing as prediction, and political discipline as public-service modernization. The next phase of digital authoritarianism may therefore not look like a return to the old police state. It may look like a smarter bureaucracy.

The literature on digital authoritarianism has rightly focused on censorship, surveillance, propaganda, facial recognition, and online manipulation. Frantz, Kendall-Taylor, and Wright (2020) show how “digital dictators” use new technologies to monitor citizens, censor dissent, and manage information. But AI shifts the emphasis from watching to administering. It does not only help regimes identify dissent after it appears. It helps them classify populations, assign risk, automate decisions, and embed control inside ordinary systems of governance; pushing the project of making citizens legible in ways that go beyond what Scott (1998) envisioned.

This is what I mean by AI-enabled administrative authoritarianism: rule through systems that appear technical, neutral, and service-oriented, but that also expand the state’s capacity to sort, monitor, and discipline society. AI enters through policing dashboards, welfare databases, traffic systems, smart-city platforms, fraud detection, border control, and predictive risk tools. These applications are not inherently authoritarian. In democracies, too, they raise serious questions about accountability, bias, and privacy. But in authoritarian or semi-authoritarian contexts, where courts are weak, media scrutiny is limited, and bureaucracies answer upward rather than outward, the same tools can become instruments of political control.

Many middle-income countries face real administrative problems: corruption, inefficient policing, uneven service delivery, weak tax capacity, and distrust of frontline officials. Unlike many low-income states, however, they often possess enough fiscal capacity, procurement infrastructure, or external financing to acquire AI systems before they have built the institutions needed to govern them. AI therefore promises a shortcut to competence. It tells citizens that the state can become faster, cleaner, and more responsive without becoming more democratic.

This is where administrative efficiency and authoritarian control begin to blur. Dictators need information, but fear and hierarchy distort what reaches them as citizens hide preferences, local officials manipulate reports, and security agencies protect their own interests. Xu argues that digital surveillance helps dictators address this information problem by making society more observable (Xu 2021). Beraja et al.’s study of “AI-tocracy” gives this claim empirical force. In China, local unrest increased government procurement of facial-recognition AI, linking political control to the routine purchasing decisions of local public-security agencies (Beraja et al. 2023, 1349, 1358). The importance of the case is not simply that China uses surveillance. It is that AI becomes embedded in the everyday administrative machinery of public order through procurement systems, policing bureaus, personnel allocation, and urban security infrastructure. Beraja et al. also show that the use of AI altered the personnel composition of local public-security agencies, suggesting that AI does not merely add a new surveillance device; it reorganizes the labor of repression itself (Beraja et al. 2023, 1376). Repression is therefore not only carried out through exceptional crackdowns or security forces. It is routinized through the bureaucratic management of risk.

This does not mean AI makes authoritarianism omnipotent. Farrell warns that AI governance may amplify rulers’ biases and blind them to information they need (Farrell 2025). Yang similarly argues that repression can damage the data environment itself: when citizens self-censor or falsify preferences, AI systems may become less accurate precisely when regimes most need truth (Yang 2025). More data does not necessarily mean better knowledge.

Nor does this mean established democracies are immune. The same administrative logic can appear in softer forms where AI is introduced through welfare systems, policing, border control, fraud detection, or public-sector automation. The difference is institutional: democracies have courts, media, elections, civil society, and privacy regimes that can contest these tools, even if imperfectly. In authoritarian and semi-authoritarian settings, those constraints are weaker, making it easier for administrative efficiency to slide into political control. As Cupac et al. warn, AI may also contribute to democratic backsliding by encouraging hyper-technocratic and paternalistic forms of governance even within democracies (Cupac et al. 2024).

AI-enabled administrative authoritarianism is potentially dangerous not because AI automatically produces more effective repression, but because it can make political control appear as administrative improvement. In settings where courts, media, civil society, and bureaucratic accountability are weak, tools introduced for policing, welfare delivery, fraud detection, or urban management may also expand the state’s capacity to classify, monitor, and discipline citizens. The political risk, then, is not only that authoritarian states become more coercive, but that coercion becomes harder to identify: repression can be reframed as risk management, surveillance as service delivery, and compliance as a condition for accessing the state.