For most of history, the state was powerful but partly blind. AI cities may end that blindness, creating a new political temptation: not the dramatic rise of dictatorship overnight, but the slow normalization of surveillance as public service.
In Jakarta, this is already concrete. In December 2025, Indonesia’s national traffic police announced a target of 1,000 electronic traffic enforcement (ETLE) cameras across the greater Jakarta area by 2026, up from 127 today, with thousands of CCTV units to be integrated. In May 2026, Korlantas announced that ETLE is being upgraded with facial recognition linked to Dukcapil, the national civil registration database, to identify offenders when vehicle data is unclear, as reported by Indonesian outlets including Detik and Kumparan.
This looks like ordinary traffic modernization, but it marks a deeper shift. Enforcement is no longer only about vehicles and violations; it is becoming connected to biometric identity and population data, and the governance question grows far larger than road safety.
The pattern is not unique to Jakarta. In London, the Metropolitan Police has expanded live facial recognition in public spaces, scanning faces against watchlists; police call it a tool for catching suspects; civil liberties groups warn it normalizes mass surveillance. The European Union has drawn the opposite lesson: its AI Act treats certain biometric and public-sector applications as high-risk or unacceptable-risk, on the premise that some AI is less an ordinary digital upgrade than a system affecting fundamental rights and democratic accountability.
These cases converge on one question: when cities become intelligent, do they become more accountable, or merely more capable of watching their citizens? This matters most in the Global South, where governments face real pressure to improve public services but often lack the safeguards to control the power AI creates. The danger is not only new surveillance infrastructure, but it is also political. Urban AI that expands without accountability can produce a gradual tech-enabled authoritarian drift: less open dictatorship than the quiet normalisation of state visibility, behavioural monitoring, and automated suspicion.
The Global South Is Entering the AI City Era
AI cities will not arrive as an ideology. They will arrive as solutions to everyday urban frustration: congestion, crime, flooding, poor services, and slow bureaucracy. A transport agency uses AI to optimize routes; a welfare agency targets assistance through analytics; a police unit uses facial recognition to identify suspects. Each use case sounds reasonable in isolation.
But connected to identity systems, CCTV networks, biometric recognition, and welfare or transport databases, these tools become something larger than modernization: an infrastructure of urban visibility. As Siegfried Zhiqiang Wu argues in The AI City (Springer, 2025), the shift is from smart cities that process information to AI cities that learn, predict, coordinate, and adapt. The AI city is not simply a smarter dashboard. It is a new way for the state to see, classify, and act.
Urban Intelligence Is State Power
In the twentieth century, strategic infrastructure meant ports, railways, energy grids, and telecommunications. In the twenty-first century, urban intelligence systems are becoming a comparable layer of state power, shaping how the city is sensed, prioritized, and governed. Integrated with public databases, AI systems can shape access to welfare, policing, housing, and mobility. AI cities are best understood as state-capacity infrastructure.
For Global South governments, that is tempting; many states genuinely need to see problems earlier and allocate resources better. But state capacity without accountability becomes state visibility without restraint. And this is not only a governance issue. It is a sovereignty issue, because few cities build these systems alone. The market for urban surveillance is shaped by great-power competition.
Chinese firms such as Huawei, Hikvision, and Dahua dominate the Global South market, often backed by concessional loans under the Digital Silk Road; the Carnegie Endowment has documented Chinese AI surveillance technology in more than sixty countries. Yet the dynamic extends well beyond China. French firms such as Idemia and Atos are major biometric players, and European export credit agencies finance comparable projects with little scrutiny. For a Global South city, the danger is less the nationality of a vendor than dependency itself: when cloud infrastructure, algorithms, and biometric databases are designed and controlled abroad, the state’s capacity to see its own citizens is partly governed from beyond its borders.
This turns AI city procurement into a quiet geopolitical choice. Competing powers export not only equipment but also governance models: one built around centralized state visibility, another that at least formally embeds rights and oversight. The European Union’s AI Act is partly an attempt to make the second kind of standard travel. The risk for the Global South is becoming a rule-taker, importing whichever model is cheapest or best-financed rather than shaping the terms on which urban intelligence is built. And once these systems become everyday infrastructure, reversing them is politically and institutionally difficult.
The Efficiency-Surveillance Trap
The most dangerous AI systems in public governance are rarely the ones that fail. They are the ones that work well enough to expand without scrutiny. This is the efficiency-surveillance trap: systems designed to improve public services expand the state’s capacity to observe, classify, and restrict citizens faster than institutions can govern that power.
A CCTV system introduced for safety becomes facial recognition infrastructure. Transport analytics become mobility profiling. Fraud detection becomes citizen risk scoring. Welfare targeting becomes automated exclusion. This rarely happens through sinister design; it happens through institutional convenience. Once data exists, new uses become tempting; once officials experience the benefits of visibility, they ask for more.
A false match in a traffic system may produce more than a fine; joined to other databases, the same error could affect policing attention or access to services. This is why Jakarta matters. A traffic system linked to facial recognition and civil registration may begin as a tool for road discipline, but if governance is weak, it can become a template for biometric enforcement across public space, policing, and services. The issue is not whether traffic enforcement is necessary; it is whether enforcement infrastructure can expand into population-level monitoring without public debate, auditability, purpose limitation, and redress. By the time such a system is fully deployed, it is operationally useful, politically convenient, and hard to reverse.
These dynamics hit the Global South hardest, where the dilemma is sharpest: how to build state capacity without unaccountable state visibility. Governance tends to lag behind technology, and surveillance is politically attractive: citizens facing crime may welcome intrusive systems, so it arrives as convenience more than as repression. Here AI does not simply modernize the state; it compounds whatever weaknesses already exist, turning patchy data into automated error and thin oversight into expanding reach. The result is a double risk: surveillance from above and a collapse of legitimacy from below if citizens encounter AI first as a tool of control.
Minimum Viable Safeguards
The answer is not to reject AI cities; Global South governments need better tools, and citizens deserve better services. It is to make governance a condition of adoption. What most institutions need is a governance floor: deliberately minimal for routine uses and demanding only where the stakes are high.
First, sort systems by stakes. The highest-stakes category, covering facial recognition, predictive policing, automated welfare eligibility, and cross-agency identity linking, should trigger stronger review before deployment.
Second, require public-sector AI registers. A state cannot govern systems it has not recorded; it must know where AI is used, for what purpose, with what data, and under whose responsibility.
Third, make traceability mandatory. Sensitive data access, model updates, high-impact decisions, and human overrides should leave an audit trail. Distributed ledger technology can help. Stripped of its speculative associations, it offers something practical: a durable record of who accessed data and who signed off on decisions. The point is to ensure that the exercise of power leaves a trail.
Fourth, build redress before scale. If an AI system can restrict access, flag suspicion, or affect public benefits, citizens need a clear path to explanation, correction, and appeal before it becomes permanent infrastructure.
Fifth, use procurement and donor conditionality as leverage. Development banks and vendors should not fund AI city projects that lack basic safeguards: auditability, data protection, human review, and public notice. In a market where rival powers export competing governance models, procurement is also where the Global South decides whether it sets its own terms or absorbs someone else’s.
Finally, make accountable AI politically rewarding, a reputational asset, and a marker of a city’s competitiveness. Reform should not depend on the goodwill of elites. It should depend on making transparency pay better than quiet control. These safeguards are not anti-innovation; they are what allow innovation to survive contact with politics.
Intelligent Cities Need Accountable Power
The twenty-first century will be decided in cities and increasingly by how those cities think. For the Global South, urban AI is less a threat to resist than a transition to govern. The tools are arriving regardless; the open question is on whose terms.
That is genuinely hard. Building accountability at the speed of capacity and resisting dependency without refusing investment is never the path of least resistance. But difficulty is not destiny. The same competition between governance models that exposes the region to dependency also hands it leverage: suppliers need buyers, and buyers can attach conditions. The cities that lead this century will not be the ones with the most sensors or the most powerful command centers. They will be the ones that make intelligence answerable and that treat the freedom of their citizens as a measure of good governance rather than a cost of it.
The state will grow more intelligent; that much is settled. Whether that intelligence comes to serve the public or merely to watch it is still a choice, and it is the Global South’s to make.
