China's AI-Powered Censorship: Supercharging Online Content Control

03/26/2025 Social Media

A leaked database reveals China is leveraging AI to enhance its online censorship capabilities, going far beyond traditional methods. The system, trained on 133,000 examples, flags content deemed sensitive by the Chinese government.

The AI system appears designed to censor Chinese citizens online and refine Chinese AI models' censorship abilities. Xiao Qiang, a researcher at UC Berkeley, confirmed the database as evidence of the Chinese government's intent to use LLMs for repression, improving state-led information control.

AI-Driven Censorship

Authoritarian regimes are rapidly adopting AI technology. OpenAI reported Chinese entities using LLMs to track anti-government posts and defame dissidents. The Chinese Embassy stated its opposition to "groundless attacks" and emphasized the importance of ethical AI development.

The dataset, found by NetAskari on an unsecured Baidu server, contains recent entries from December 2024. The system tasks LLMs to identify content related to sensitive topics, marking "highest priority" items for immediate attention.

Targeted Content

Prioritized topics include pollution scandals, food safety issues, financial fraud, and labor disputes. Political satire and mentions of Taiwan politics are explicitly targeted. The dataset extensively covers military matters, including reports of military movements and weaponry.

The 133,000 examples include posts about corrupt police, rural poverty, and officials expelled for corruption and "superstitions." Commentary on Taiwan's military and Chinese jet fighters are also present.

Subtle dissent, like anecdotes about the fleeting nature of power, is also targeted. The dataset is intended for "public opinion work," suggesting its use by the Cyberspace Administration of China (CAC) for censorship and propaganda.

The ultimate goal is to safeguard Chinese government narratives online while eliminating alternative viewpoints. President Xi Jinping has described the internet as the "frontline" of the CCP’s “public opinion work.”

OpenAI's report showed similar AI use to monitor social media and target dissidents. While China has traditionally used basic algorithms, AI tech like LLMs enables more efficient detection of subtle criticism. Xiao Qiang emphasizes the evolving sophistication of AI-driven censorship in China.