Evidence has emerged that China attempted to develop an artificial intelligence (AI) technology akin to the film ‘Minority Report,’ preemptively identifying individuals likely to criticize the government or participate in anti-government protests. This surveillance infrastructure aggregates past and present digital traces left by citizens—such as communication records, internet usage history, social media activity, and location data—to calculate the probability of someone becoming a dissident.
The New York Times (NYT) reported on the 1st (local time) that Chinese surveillance firm Geedge Networks researched AI-based identification of politically risky individuals with government-backed research organization MESA Lab starting in early 2024, based on an analysis of over 100,000 leaked documents and Vanderbilt University researchers.
Geedge Networks previously sold a commercial version of the Great Firewall, China’s national internet censorship system. It mainly sold software to block access to specific sites or prevent the use of virtual private networks (VPNs) that bypass censorship. The company’s chief scientist is Professor Fang Binxing, dubbed the “father of Chinese internet censorship.” InterSecLab, a security research institute that analyzed the leaked documents, explained that the company supplies systems capable of deep packet inspection (DPI), which examines user data packet contents, and real-time monitoring of mobile subscribers.
Geedge Networks used data collected through its surveillance network to estimate who might pose political risks in the future. Its system integrates scattered digital data to track user behavior patterns. It first links real-name registered mobile phone numbers to messenger accounts. This is combined with device identifiers, IP addresses, and VPN bypass records to build individual profiles. Adding base station communication logs and GPS signals completes a timeline of a person’s movements, revealing when they passed protest sites or interacted with activists. Experts noted that such integration is easier in environments like China, where the state monopolizes communication networks and enforces platform cooperation.
Individuals who frequently use VPNs, primarily visit overseas media sites, repeatedly visit past protest locations, or interact with dissidents are flagged as “potential political risks.” The system can link purchases of specific books or movie tickets to movement patterns. Brett Benson, a political science professor at Vanderbilt University, stated, “Ordinary data becomes material for determining who a user is and what they might do next.”
According to the leaked documents, researchers discussed identifying people’s “intentions” and detecting “harmful information” in a February 5, 2024, meeting. Under the Chinese Communist Party’s framework, “harmful information” extends beyond serious crime data to include political dissent, regime criticism, and sensitive debates the authorities seek to suppress. Jimmy Goodrich, a senior researcher at the University of California’s Global Conflict and Cooperation Research Institute, said, “Chinese security agencies are overwhelmed by data,” adding, “The true value of AI lies in filtering core threats from massive data.” This suggests the intent to let AI pre-screen targets, as direct surveillance of 1.4 billion citizens is impossible.
Experts warned that AI-assigned risk scores could blacklist citizens without actual protest involvement due to algorithmic errors. The European Union (EU) explicitly prohibits systems that predict or assess crime risks based solely on personal traits or characteristics through its AI Act. Brett Goldstein, director of Vanderbilt’s Wicked Problems Lab, said, “This is what happens when mass surveillance meets AI,” adding, “What China is doing to its citizens is a preview of what could occur in any nation deploying unregulated AI.”
It remains unknown whether China has completed and deployed this technology. Experts speculated that China might lack high-performance semiconductors to run AI prediction models at this scale. Text-based keyword filtering via internet censorship requires minimal computational power. However, predicting future actions by linking nationwide call recordings, surveillance footage, and location data demands massive GPU infrastructure. Vanderbilt University noted in publicly available documents that Geedge Networks faced GPU limitations, resorting to older, lower-performance AI models and chips. This supports analyses that the Biden administration’s semiconductor export controls slowed China’s advancement of a high-tech surveillance state.
Even after President Trump’s inauguration, China still cannot import top-tier NVIDIA chips due to U.S. sanctions. The NYT cited U.S. officials stating, “While Geedge Networks secured GPUs for current products, implementing its most ambitious prediction systems likely requires high-performance chips that China struggles to obtain.”
Originally written by: You Jin-woo
Image credit: Yonhap
Source: The Chosun Daily
Published on: 2 June 2026
Link to original article: China’s AI System Preemptively Identifies Potential Dissidents
