China’s AI predicts dissent before citizens speak, leak reveals

A bombshell New York Times investigation, analyzed by Vanderbilt University researchers, has exposed a leaked dataset from Beijing-based tech firm Geedge Networks. The documents reveal that China is actively developing AI-driven predictive surveillance systems that go far beyond traditional internet censorship. These systems leverage large language models to aggregate real-time browsing histories, cell tower location data, and social media connections. By synthesizing this data at scale, they generate comprehensive citizen profiles and assign political risk scores — flagging individuals who might become critics of the government based on inferred intent, not actual actions. This marks a terrifying evolution from punishing dissent to predicting it before it happens.

The leaked files show that Geedge’s flagship product, effectively a “Great Firewall in a box,” has already been exported to authoritarian regimes aligned with Beijing’s Belt and Road Initiative. In Myanmar, the military junta used it to locate pro-democracy activists, block social media, and trigger regional internet blackouts leading to targeted arrests. Similar mass surveillance deployments with deep packet inspection and reputation scoring have been uncovered in Pakistan and Kazakhstan. Crucially, the leak also reveals a critical vulnerability: United States export controls on advanced semiconductors have successfully starved Geedge of the high-end computing power needed to scale these predictive AI models. Forced to pivot to less efficient hardware due to chip shortages, their progress has been significantly slowed, underscoring that maintaining tight semiconductor sanctions is the primary line of defense against this predictive surveillance grid expanding globally.