China is modernizing its nationwide surveillance system by integrating AI vision and language models from Hikvision and Huawei, enabling automated behavior analysis and text-based video search. The upgrades, detailed in a Financial Times report, follow a 2024 directive from Public Security Minister Wang Xiaohong after a series of violent attacks linked to a mental health crisis exacerbated by pandemic lockdowns and economic stagnation. Local authorities are equipping millions of old cameras with computer vision and language models, shifting the system from reactive identification to large-scale behavioral monitoring. According to the report, these systems are designed to detect erratic driving, crowds forming, unauthorized access, or suicidal behavior on bridges and trigger alerts automatically.
Hikvision's latest generation allows officers to search footage by typing text queries, such as looking for a woman wearing a red hat. A Hikvision manager told the FT that police no longer need to review footage manually, as the system can find clips based on text prompts. Hikvision itself said its products digitize routine tasks that previously relied heavily on manual review. Procurement documents mention projects like Yaodu's 175 HD cameras with smart video analysis and Datong's police tender for Hikvision cameras that identify gender, posture, and clothing.
Early deployments focus on dense urban areas and zones around military and government buildings. Other agencies are replacing intermediate servers with AI PCs to process video locally and reduce cloud costs. Rights experts warn that the philosophy behind China's surveillance system is becoming more sweeping. Maya Wang of Human Rights Watch cautioned that AI and computer vision give authorities an unprecedented capacity to monitor behavior at scale.
In a recent policy paper, Anthropic warned that if the compute gap keeps closing, China could not only catch up technologically by 2028 but also scale AI-powered surveillance and repression.
Source: thedecoder
