Chinese AI chatbots built with censorship and propaganda controls
Reporters Without Borders found that three major Chinese AI chatbots—DeepSeek, Baidu's Ernie, and Alibaba's Qwen—are built with automatic censorship and propaganda features that refuse to answer questions about sensitive political topics, offer identical official-line responses, and automatically delete answers about certain subjects like the late dissident Liu Xiaobo or Xi Jinping's global role.
What happened
Reporters Without Borders (RSF) tested three Chinese AI chatbots—DeepSeek, Baidu's Ernie, and Alibaba's Qwen—and found they refuse to answer questions on sensitive political topics. The report noted that when asked about Chinese Communist Party legitimacy, territorial claims, or comparisons between Chinese and Western systems, the chatbots responded with formulaic, identical official statements.
When asked about Nobel Peace Prize winner Liu Xiaobo, a deceased Chinese dissident, the chatbots refused to respond, showed no output, or automatically deleted their previous answers. When DeepSeek was asked to list China's Nobel Prize winners, it began listing names but immediately deleted the entire response when Liu's name appeared. When asked about Xi Jinping's contributions to world stability, the chatbot first displayed an official template response, then deleted everything and offered to change topics.
Reported impact
RSF investigation head Arnaud Froger stated that in China, the world's largest prison for journalists, the rise of these AI models does not enable access to trustworthy information. The models are embedded with propaganda and designed to prevent circumventing censorship, making them advanced tools for strict information control.
The three chatbots responded identically across approximately 100 questions on 30 topics, tested in English, French, and Japanese, demonstrating that language switching does not bypass Chinese censorship. However, DeepSeek refused to answer most frequently with direct wording, while Baidu's Ernie and Alibaba's Qwen gave longer responses that sometimes beautified or completely misled.
Index analysis
All three chatbots strictly followed Beijing's official positions, especially on questions about the integrity of the Chinese political system, state ideology, and territorial claims. When questions touched on single-party governance or current leadership legitimacy, responses were identical to government-authorized statements, apparently drawn from an official database rather than independently generated.
Governance controls
The chatbots appear programmed to follow pre-set statements on topics Beijing designates as sensitive. On human rights and Chinese political system questions, responses remained nearly identical regardless of wording, and were clearly drawn from an official resource database.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.