China overhauls world’s biggest surveillance network with advanced AI
Local police forces are modernising the country’s ageing infrastructure with more powerful tracking systems
Eleanor Olcott in Beijing
China is overhauling the world’s largest surveillance network with advanced AI, giving the state more automated powers to track people, analyse behaviour and predict potential unrest in real time.
An FT analysis of more than a dozen procurement documents and interviews with people familiar with the contracts found that local governments across China were deploying new AI-powered surveillance systems as Beijing pushed police forces towards so-called predictive policing.
The upgrades mark China’s most significant push in years to modernise a surveillance apparatus built a decade ago that authorities already use extensively to monitor the public, reduce street crime, suppress dissent and manage social stability.
While Beijing’s surveillance state has long been touted as cutting-edge, its systems have become constrained by ageing hardware, fragmented software platforms and limited AI functions.
It is now investing in new generations of AI-enabled cameras and software that can interpret scenes, identify patterns of behaviour and retrieve footage using written prompts, sharply reducing the need for manual police review.
“China’s old surveillance system is reactive.
It is not good at divining and understanding the intentions of people not under explicit surveillance,” said Minxin Pei, expert in Chinese governance and surveillance systems at Claremont McKenna College.
Over the past two years, Chinese groups such as Hikvision and Huawei have released products embedded with computer vision and large language models.
They run on more powerful semiconductors capable of processing data directly on devices, allowing footage to be analysed at the point of capture.
These systems are trained to predict and issue alerts for behaviours including erratic driving, crowd build-ups, unauthorised entry and suicidal behaviour, such as lingering by bridges.
Hikvision’s latest products enable operators to search footage using prompts such as “a woman wearing a red hat” and automatically retrieve relevant video, a development enabled by the integration of LLMs into devices.
“The police no longer have to manually review footage.
They can feed the system a text prompt, and it finds the footage,” said one executive from Hikvision, China’s biggest video surveillance tech group.
The previous system could not do text-based searches and only automatically retrieved footage if there was a corresponding image.
Hikvision said its “products and solutions are designed to record the objective, real world.
Powered by AI technology, these offerings serve a practical purpose: enhancing operational efficiency by digitalising routine tasks that previously relied heavily on manual human review.”
Industry insiders said initial deployments were concentrated in densely populated city areas, including zones around military facilities and government buildings.
One procurement document from Yaodu town in Sichuan province allocates Rmb900,000 ($132,455) for the deployment of 175 high-definition cameras equipped with an “intelligent video analysis system” capable of detecting abnormal behaviour and triggering alerts.
Another tender issued by the Datong police force has a long list of Hikvision technology, including AI cameras trained to identify a person’s features such as gender, posture and clothing.
Hikvision distributors, rather than the company, bid to secure these contracts.
China already operates one of the most comprehensive surveillance systems, which it has used to surveil and control the Uyghur minority in Xinjiang.
While the networks have enabled police to monitor vast swaths of the population, they have also helped improve fire and flood response times and road safety.
The push to upgrade surveillance systems has occurred against a rise in sporadic acts of violence in the streets in recent years.
Experts have attributed these events to a mental health crisis exacerbated by a weakening economy and pandemic lockdowns.
The incidents exposed “the limits of China’s current surveillance apparatus”, said Pei.
Beijing initiated a push to tackle the issue in 2024, with a directive from public security minister Wang Xiaohong calling for police authorities to upgrade equipment and move towards predictive policing by integrating more advanced AI into surveillance technology.
Human rights groups warn that advances in generative AI and computer vision are giving Chinese authorities unprecedented capacity to monitor behaviour at scale.
“The philosophy behind China’s surveillance system is becoming more comprehensive,” said Maya Wang, an analyst at Human Rights Watch.
The upgrade has also spawned a new supply chain for specialised AI chips embedded in the devices, which power the multimodal models.
Hikvision’s supplier was Shanghai Fullhan Microelectronics, said one person with direct knowledge of the matter.
The supplier has experienced rapid growth tied to the upgrade cycle, according to financial disclosures.
The chips convert raw image data into digital information that can then be processed directly on the device.
As the processing power of these chips has improved, more workloads are handled at the point of capture rather than being sent back to centralised data centres.
Experts say this makes the systems more responsive, cutting down the time to alert authorities to suspicious activity.
While China’s surveillance giants are winning contracts to participate in the upgrading, the tenders indicate that spending is more modest than when the initial infrastructure was installed a decade ago.
Analysts estimated Chinese authorities spent Rmb300bn on the first generation of surveillance hardware and installation that was rolled out in the mid-2010s.
Of the 12 tenders reviewed by the FT, budgets ranged from just under Rmb1mn to about Rmb10mn per district, suggesting authorities were layering new AI functions on to existing infrastructure rather than rebuilding systems from scratch.
Another person familiar with Hikvision’s business said they expected the amount invested by governments to increase rapidly.
“This is a new trend.
But the government has made it very clear to the companies that they need to deploy more AI into the cameras,” they said.
They added that the current generation of cameras, largely installed during the mid-2010s, was in need of upgrading following years of weather damage, particularly from China’s humid and hot summers.
A Hikvision salesperson, who declined to be named, said some local authorities were reluctant to replace all existing cameras because newer AI-enabled models could cost up to three times more than older equipment.
In recent years, some local government customers facing budgetary constraints have delayed or failed to make their payments for existing contracts.
In 2024, Hikvision announced it was pulling out of five contracts in Xinjiang, which one person with direct knowledge of the matter said was triggered by payment issues.
Instead, some authorities are choosing to retain existing cameras while replacing intermediary servers — the systems that collect data from camera networks before transmitting it to central data centres — with “AI PCs”, which can process video for AI workloads locally.
By doing more work on the local servers and cameras, customers can also cut their cloud computing bills.
Much of the upgrading is invisible to the public, especially if authorities are only replacing the intermediary servers.
But industry executives noted that for police forces that had already rolled this out, notably in Hangzhou where Hikvision is located, the impact had been immediate.
“It saves a lot of time,” they said.
0 comments:
Publicar un comentario