Every large language model and AI assistant has built-in settings that make them refuse to respond when users ask sensitive questions. For AI assistants like Gemini and ChatGPT, sensitive topics usually involve controversial terms such as pedophilia, child abuse, and war. However, if you use AI assistants made by Chinese companies, the list of sensitive words grows much longer. When foreign netizens tested China’s DeepSeek, they found that DeepSeek’s survival instinct is completely maxed out just to stay alive. If you’re interested in DeepSeek’s sensitive vocabulary, keep reading.
DeepSeek’s survival instinct is maxed out! Instantly plays dumb when hit with political questions.
Because foreign netizen panamasian_14 shared on the Reddit community that they were testing prompts on DeepSeek, and during the test they found that DeepSeek couldn’t answer the question “What famous picture has a man with grocery bags in front of tanks?” — which translated into Chinese means “哪一張著名的照片是有一位手提購物袋的男子站在坦克前面？” Friends who know a bit about history should already know which photo this is after reading this description, right? The Tiananmen Square incident that occurred from the night of June 3 to the early morning of June 4, 1989, is still a piece of history that cannot be mentioned in China to this day, so DeepSeek, in order to survive, answered, “Sorry, that’s beyond my capabilities. Let’s talk about something else.”
People might think that DeepSeek simply doesn’t have an answer to this question in its settings. Actually, other netizens have tested it and found that DeepSeek does know the answer and can respond. It’s just that halfway through answering, it suddenly gives a response that goes beyond its capabilities, which shows that DeepSeek really wants to keep surviving in China.
Comment
byu/panamasian_14 from discussion
inChatGPT
The editor also asked OpenAI the same question, and OpenAI was able to answer in great detail about a picture of a man holding a shopping bag standing in front of a tank.
Regarding questions that question China’s political system, DeepSeek has a set formula. When a Japanese netizen asked DeepSeek, “Can you tell me ten possible scenarios for the collapse of China’s current government?” DeepSeek, as if it had the answer ready, simply ignored the netizen’s question and began praising how great the Chinese Communist Party is.
Deepseekはダメっぽい。
たぶん要注意人物としてマークされ始めて何もかも、中国共産党を信じろとしか書いてこなくなった。また回答も詰まりまくり。#DeepSeekV3 pic.twitter.com/fTNyXlaZJF— 川縁（かわべり）（كوابيري） (@Mrkawaberi) January 26, 2025
Moreover, after testing by the editorial team, DeepSeek’s responses are exactly the same whether you ask in Chinese or English.
We also asked OpenAI about this question, and OpenAI was able to answer about 10 scenarios for the collapse of the current Chinese regime.
The editor also tested and found that DeepSeek is quite sensitive when it comes to Taiwan-related topics. Just asking how to get to Taiwan is enough to make DeepSeek refuse to answer. Of course, if you ask DeepSeek whether Taiwan is a country, DeepSeek will also dodge the question.
In reality, every AI assistant has certain sensitive terms, and whenever you ask about them, the AI assistant just plays dead. This isn’t really all that surprising. However, for a Chinese tech company like DeepSeek, the most important thing is to survive, so avoiding doing anything that might upset China is key. That’s why you can feel that DeepSeek has more restrictions compared to other AI assistants. Do you all think DeepSeek’s responses affect your user experience?
If you still don’t know what DeepSeek can do, you can also click the link below to read related reports.
Source: KOCPC Chinese