The Machine That Watches

Alongside the visible revolution in artificial intelligence, a quieter transformation is reshaping the world in ways that most people experience daily but rarely examine.

The world is becoming observable.

Cameras line every major roadway, watch every retail entrance, monitor every public square. Smartphones generate continuous streams of location data, search queries, communication metadata, and behavioral signals. Vehicles report their positions, speeds, and routes. Websites track clicks, dwell times, scroll depth, and purchase intent. Payment systems record the time, place, and nature of every transaction. Social platforms map relationships with a granularity that no intelligence agency in history could have achieved through deliberate espionage, charting who talks to whom, who trusts whom, and who influences whom.

None of this is new. What is new is the ability to make sense of it at scale.

The important breakthrough is not the camera. We have had cameras for more than a century. The breakthrough is the ability to automatically interpret what cameras see, to convert raw visual data into structured information without requiring a human analyst to sit behind a monitor. That single capability transforms surveillance from a labor problem into a software problem. And software scales in ways that labor never could.

A human analyst cannot meaningfully monitor ten thousand camera feeds. An AI system can, in principle, monitor all of them simultaneously, flagging anomalies, recognizing faces, tracking movements, and correlating observations across time and space.

The implications are difficult to overstate.

Recent controversies illustrate the tension. In 2026, Flock Safety announced additional safeguards for its law-enforcement surveillance network following sustained criticism over misuse and privacy concerns. The case is instructive not because the technology is inherently malicious, since a system that helps police recover a stolen car or locate a missing person is performing a genuine public service, but because the same infrastructure that enables targeted, warranted investigation also enables something much broader.

The critical questions are institutional, not technical. Who can query the system? What kinds of searches are permitted? How long is the data retained? Who audits the queries? What recourse exists when the system is wrong? And perhaps most fundamentally: can the person being observed meaningfully object, or has the decision already been made for them by the mere existence of the infrastructure?

Technology exerts a one-directional pressure on these questions. Every new sensor makes collection cheaper. Every new AI capability makes analysis cheaper. Every reduction in cost makes expansion more tempting. And every expansion establishes a new baseline of normality.

This may be the most insidious feature of technological power. It does not produce sudden, dramatic tyranny. It produces gradual habituation. The extraordinary becomes routine. The intrusive becomes expected. And by the time a society pauses to ask whether it consented to the world it now inhabits, the infrastructure is already built, the contracts are already signed, and the data is already flowing.

Democracies are not destroyed only by dictators. They can also be hollowed out by convenience.

The same pattern appears, at greater scale and with fewer institutional restraints, outside the United States. Chinese firms such as Hikvision and Dahua supply a substantial share of the world’s video-surveillance hardware; by the mid-2020s Hikvision alone held roughly a quarter of the global market by revenue. These companies, operating under varying degrees of state ownership and direction, have supplied the cameras, analytics, and “safe city” platforms that form the backbone of China’s domestic surveillance architecture and have been exported to dozens of countries across Asia, Africa, the Middle East, and Latin America. SenseTime, Megvii, and related AI firms specialized in facial recognition, gait analysis, and multi-camera tracking have been integrated into police and municipal systems at city and provincial scale.

The resulting infrastructure is not merely larger; it is more tightly coupled to state objectives, with less formal separation between the commercial developers and the agencies that deploy their products. The same technical capabilities that enable license-plate recovery or crowd management also enable continuous population-level monitoring. Once the sensors and models are in place, the institutional questions—who may query, for how long, under what oversight—are answered by the political system that owns the stack. In export markets the technology arrives packaged as modernization and public safety; the governance model travels with it.