700 million cameras and an unresolved dilemma: how AI turns Chinese surveillance into an uncomfortable mirror for the rest of the world

By Zendoric · June 28, 2026.

The figure is hard to grasp on a human scale: 700 million cameras deployed across China's cities, roads, stations and airports. The economist José Ramón Riera translated it into a more comprehensible proportion during a recent public appearance: one camera for every two inhabitants, more devices installed than all of Europe has inhabitants. But the number, on its own, would be anecdotal without what turns it into a system qualitatively different from any previous video-surveillance network: the integration of artificial intelligence capable of identifying people, reconstructing movements and analyzing behavior in practically real time.

What China has built is not an expanded version of the closed-circuit television we know in shopping malls or airports. It is an infrastructure where facial recognition, automated image analysis and the massive cross-referencing of data turn each camera into a node of a continuous cognitive network. The Chinese authorities argue that the system reduces crime, improves public safety and optimizes the management of large cities. The problem is that operational efficiency and the safeguarding of rights are not the same thing, and history shows they rarely go hand in hand without explicit regulation.

The most relevant aspect of the debate, however, is no longer China. As Riera noted, «this is no longer only happening in China». The facial-recognition systems that were born in U.S. laboratories have been replicated in dozens of countries, including European ones; airports across half of Europe use biometric readings routinely. The difference between the Chinese model and the Western models is not technological but regulatory: the European General Data Protection Regulation, the EU's recent AI Act —which bans real-time biometric identification in public spaces except in narrowly defined cases— and the existence of independent judicial systems establish a perimeter of control that in China simply does not exist in the same way. But that perimeter is porous and under constant pressure: governments of various stripes in Europe have in recent years requested an expansion of police use of facial recognition, and every attack or security crisis renews that impulse.

Riera put it in clear political terms: «every time there is a government that wants to keep us more controlled, there will be more cameras». It is an observation that goes beyond the Chinese regime. The demand for surveillance is neither ideologically neutral nor geographically localized; it is a response to perceived insecurity that can emerge in any political system when social pressure exceeds the threshold of risk tolerance. AI tools have enormously cheapened the marginal cost of each new observation node: adding a smart camera to an existing network costs a fraction of what it cost a decade ago, and its analytical performance has improved exponentially.

This raises a forward-looking question that democracies can no longer keep postponing: what governance model do we want for the algorithmic surveillance of public space? The Orwellian reference Riera used is, in a way, too comfortable: George Orwell's 1984 described a State that actively watched in order to repress dissent. What we are living through is more ambiguous and, for that reason, harder to combat: systems installed on legitimate security grounds, whose use can drift toward political control without any single moment being clearly the crossing of the red line.

As context for the sector, it is worth noting that the global smart video-surveillance industry moves tens of billions of dollars a year, with Chinese players such as Hikvision and Dahua among the world's largest hardware exporters. Europe and the United States have begun to restrict the purchase of these companies' equipment for critical infrastructure, but their prior penetration of municipal and private markets is already considerable. The surveillance supply chain is, in practice, globalized.

The end point is not necessarily pessimistic. Biometric-recognition technology can have genuinely beneficial uses: locating missing persons, identifying suspects in criminal investigations with judicial safeguards, managing traffic flows without involving personal data. The challenge lies in building regulatory frameworks that distinguish those uses from indiscriminate mass surveillance, and in equipping them with independent oversight mechanisms that have real teeth. The European AI Act is a first step; its effective application, still to be consolidated, will tell whether it serves as a firewall or as a dead letter.

What the Chinese case puts on the table, in short, is not a distant external threat. It is the clearest available mirror of how far any society can go when surveillance AI scales without democratic limits. Looking at it coldly, neither demonizing the technology nor ignoring its implications, is the most urgent intellectual and political exercise we have before us.