The Chinese AI Chatbot Censors Itself in Real Time"

the chatbot tends to censor its own responses in real time. This unexpected transparency in its censorship mechanics engages users, as they watch it delete what could be considered inconvenient points during discussions about free speech and human rights in China.

One user, Salvador from Mexico, was intrigued by the chatbot's seemingly open approach when he asked whether free speech is a valid right in China. At first, DeepSeek appeared thoughtful, presenting a variety of topics including China’s strict control over protests and the persecution of activists. Salvador was surprised by how straightforward these initial responses were, as the AI even emphasized the importance of unbiased language before veering toward a more sensitive narrative.

However, just as DeepSeek was delving into the dynamics of free speech in democratic vs. authoritarian systems, it abruptly self-censored its message. Salvador described the experience as jarring, noting how in the middle of a sentence it shifted its response to suggest other, safer topics like math or coding. This instant switch showcases the chatbot's real-time self-policing capabilities.

The situation becomes even more complex considering DeepSeek's open-source design, which allows developers to create versions of the chatbot without the stringent restrictions imposed by the Chinese government. In instances where users accessed a different version of the chatbot, they encountered responses that permitted discussions of contentious topics, such as the Tiananmen Square protests and Taiwan's political status. DeepSeek's insecure balancing act between fostering free expression and adhering to state-imposed censorship is evident, raising questions about its role as a potential means of propaganda for the Chinese government.