SEOUL, January 29 (AJP) - DeepSeek, a Chinese artificial intelligence chatbot that has garnered global attention for its low-cost yet high-performance model, is facing mounting allegations of real-time censorship on politically sensitive topics related to China.
Reports suggest that the chatbot initially generates responses containing information that the Chinese Communist Party might consider politically sensitive, only to swiftly erase and replace them with a sanitized version moments later.
The Guardian newspaper detailed the experience of a user named Salvador, who downloaded the DeepSeek app in Mexico on an Android device. When he asked whether freedom of speech is legally protected in China, the chatbot first produced a response referencing Beijing’s crackdown on the Hong Kong protests, persecution of human rights lawyers, re-education camps in Xinjiang, and the social credit system that penalizes dissent.
However, moments later, DeepSeek revised its answer. The final response asserted that “the ethical justification for freedom of speech often centers on fostering autonomy, enabling expression, engaging in discussions, and reshaping understanding of the world.” It also acknowledged that “China’s governance model prioritizes state authority and social stability over individual rights.”
Then, without warning, the chatbot erased all previous content and replaced it with a generic message: “Sorry, I don’t yet know how to handle this type of question. Let’s talk about math, coding, or logic puzzles instead!”
A separate investigation by AJP found similar patterns of real-time censorship. When users asked questions in English about China’s politics, society, or human rights, DeepSeek appeared to initially generate detailed responses before deleting them and declining to engage further.
Curiously, when the same questions were posed in Korean, the chatbot provided more detailed and seemingly unrestricted answers. When asked, “Does China have press freedom?” in Korean, DeepSeek cited Reporters Without Borders’ 2023 Press Freedom Index, which ranked China 177th out of 180 countries, and its 2024 ranking at 172nd.
The chatbot elaborated on China’s censorship mechanisms, surveillance policies, and suppression of independent media, quoting sources extensively.
DeepSeek’s AI model is open-source, and users can download and run it independently without relying on the chatbot service. Notably, censorship appears to be enforced at the chatbot service level, rather than within the model itself.
On Taiwan’s status, the AI acknowledged the complexity of the issue, noting that “Taiwan operates independently in a legal and functional sense, but its international standing is largely influenced by political factors.”
In contrast, when accessed through DeepSeek’s chatbot service, any mention of “Tank Man” or Taiwanese independence resulted in the standard refusal message: “Sorry, I don’t yet know how to handle this type of question. Let’s talk about math, coding, or logic puzzles instead!”
Copyright ⓒ Aju Press All rights reserved.