China uses artificial intelligence to transform its surveillance system into a tool of "predictive" control

CCTV cameras on a street in a Chinese city. Photo: Witthaya Prasongsin / Getty Images

FT journalists analyzed more than a dozen government tenders and spoke with people familiar with the contract details. They concluded that local authorities across China are actively implementing new AI-powered surveillance systems.

China already possesses one of the most comprehensive surveillance systems in the world. As the FT reminds, it has already been used to control the Uyghur population in Xinjiang. But, as experts note, the previous surveillance network was more reactive. It worked well once a person was already under surveillance but poorly at recognizing the intentions of people not yet subject to overt control.

As the FT writes, the main goal of the modernization is a transition to so-called "predictive policing." While the old infrastructure, created a decade ago, was passive and required police to spend huge amounts of time manually reviewing videos, the new systems work proactively.

The modernization of the surveillance apparatus comes amidst a rise in spontaneous acts of street violence, which experts link to a mental health crisis exacerbated by economic problems and quarantine restrictions during the pandemic. In 2024, Minister of Public Security Wang Xiaohong issued a directive to the police on the necessity of implementing AI for crime prevention.

The technological aspect of the issue is based on the integration of large language models and computer vision into products from giants like Hikvision and Huawei. New systems run on more powerful chips and can process data directly within the camera itself or on local equipment — without constant transmission of information to central data centers.

These systems are trained to predict and issue warnings about behaviors such as sudden maneuvers on the road, large unauthorized gatherings of people, illegal trespassing, or even signs of suicidal intent, if a person stays near a bridge for a long time.

Hikvision's latest developments allow operators to search video recordings using short text descriptions, such as "woman in a red hat." The system automatically finds corresponding footage thanks to the integration of large language models.

As a Hikvision manager told the FT, police no longer need to spend hours manually reviewing recordings — it's enough to enter a text query. In older systems, such a search was only possible with a sample image.

Financial Aspect of Modernization

The first large-scale implementations of new systems are taking place in densely populated urban areas, as well as around military and government facilities.

One tender in Yaodu township in Sichuan province provides for the installation of 175 high-definition cameras with an "intelligent video analysis system" capable of detecting anomalous behavior and triggering alarms. Another tender, published by Datun city police, included a large list of Hikvision equipment, including AI cameras that can determine a person's gender, posture, and clothing.

At the same time, today's expenditures appear significantly more modest than during the construction of the first wave of systems in the mid-2010s. Back then, analysts estimated that Chinese authorities spent about $44 billion on cameras and infrastructure.

Now, the budgets for the 12 tenders analyzed by the FT range from less than $145,000 to approximately $1.45 million per district.

An additional factor necessitating modernization was the physical wear and tear of equipment: cameras installed in the mid-2010s suffered greatly from weather conditions, especially the humid and hot summers characteristic of many regions of China.

However, the financial problems of local authorities and the high cost of new cameras (which can be three times more expensive than older models) are forcing officials to seek compromises.

Instead of a complete replacement of the camera fleet, many authorities are now installing "AI-PCs" — intermediate servers that process video locally. This allows not only for new features but also reduces cloud computing costs.

For most people, the modernization of the surveillance system remains almost imperceptible, especially if only servers are being updated. But, as industry representatives note, in cities where systems have already been updated — for example, in Hangzhou, where Hikvision is based — the effect was almost immediately tangible. According to them, new technologies "save a huge amount of time."