Leaked documents reveal that Chinese surveillance firm Geedge Networks, in collaboration with government-backed researchers, has been developing AI technology designed to predict political dissent by analyzing vast amounts of citizen data. The system aims to assign “risk scores” to individuals based on their digital footprint, including communication records, internet usage, location data, and social associations.
Key Points
- The surveillance system integrates diverse data points—such as VPN usage, foreign news consumption, travel history, and activist associations—to calculate the probability of a citizen engaging in anti-government activities.
- Geedge Networks, led by “father of China’s internet censorship” Fang Binxing, uses deep packet inspection and mobile monitoring to consolidate real-time user data for predictive profiling.
- Researchers and security organs intended to use the AI to manage “data overload,” allowing authorities to identify and suppress dissent without the need for manual, mass-scale monitoring.
- Legal experts and international observers warn that this technology could lead to the blacklisting of innocent citizens based on algorithmic errors, mirroring concerns about predictive policing.
- Development of the most ambitious versions of this technology has been significantly slowed by U.S. sanctions, which restrict China’s access to the advanced GPUs required to process population-wide behavior data.
Geedge Networks integrates a wide range of personal data to calculate the “political risk score” of Chinese citizens by aggregating and merging information from various digital and physical spaces.
The specific categories of personal data being integrated include:
- Communication Data:
- Real-name mobile phone numbers.
- Messenger account links.
- Call records.
- Wiretap audio from phone calls.
- Digital Footprints and Online Activity:
- Internet browsing history (specifically monitoring viewing habits, such as visiting foreign news sites).
- Social media activity.
- VPN usage history (to track attempts to circumvent censorship).
- Unique device identifiers and IP addresses.
- Physical Movement and Interaction Tracking:
- GPS signals.
- Base station records.
- Surveillance video data.
- Time-series trajectories used to map when an individual passes by a protest site and identifying which activists they have had contact with.
- Miscellaneous Personal Consumption/Behavioral Data:
- Records of books purchased.
- Records of movie theater attendance.
The system functions by linking these disparate data points to create a comprehensive user profile, which is then analyzed to determine the probability that an individual will engage in political dissent or anti-government protests.
While this initiative raises significant concerns regarding human rights and the potential for preemptive blacklisting, experts suggest that U.S.-led semiconductor export controls have currently hindered China’s ability to acquire the high-performance computing power necessary to fully deploy such a massive, real-time predictive model.
