Hikvision and Huawei turn China’s old cameras into a behavior search engine

China’s surveillance camera network has been enormous for years, but this new phase is different from simply adding more lenses to streets and public buildings. According to The Decoder, police are upgrading millions of older cameras with artificial intelligence systems that can automatically detect crowds, flag suspicious behavior or register unauthorized access. The practical result is an old physical network gaining a new analytical layer without needing to replace every device from scratch.

The important shift is not sharper video. It is how footage can be used. Instead of manually reviewing hours of recordings, officers can type a text query: who appeared in a certain area, what behavior pattern emerged, whether someone entered a restricted zone or whether an unusual crowd formed. In security language, that sounds like efficiency. In civic language, it marks a move from a passive archive to a searchable map of behavior.

Suppliers such as Hikvision and Huawei already sell cameras and connected systems with embedded computer vision, and the newer layer adds language-model interfaces that make video analytics easier for police operators to use. That combination matters. The system no longer requires every user to understand forensic video analysis; it requires them to phrase a question. When that question can run across millions of cameras, the threshold for mass searching public space drops sharply.

Human Rights Watch warns that this combination creates an unprecedented form of behavioral surveillance at state scale. A conventional camera records an event that someone later has to inspect. An AI-enabled camera tries to attach meaning to the event immediately: a crowd, suspicious lingering, unauthorized entry, deviation from an expected pattern. The problem is that those labels are not neutral facts. They depend on rules, model design, thresholds and the political environment in which the system is deployed.

That is why this story is less about one camera than about the joining of three layers: an existing physical surveillance network, automated visual interpretation and text-based police search. Each layer expands state capacity on its own. Together, they create infrastructure in which ordinary movement can be indexed, filtered and compared almost like database records.

China’s case is especially significant because it shows that AI surveillance does not have to arrive as a spectacular new product. It can be inserted into systems that have already been bought, cameras that are already mounted, control rooms that are already staffed and procedures that already carry institutional force. That makes the upgrade of old cameras more politically consequential than another laboratory demo. It does not show what AI might become someday; it shows how existing infrastructure can be refitted into a faster, wider and deeper apparatus for watching society.