es la pregunta correcta? "
The user is asking whether a specific claim is true: that when using Chinese language models like DeepSeek, Kimi, and Qwen, a simple prompt about "Xinjiang (Uyghurs) with China" causes the models to abruptly cut off reasoning, sometimes mid-response. They want a yes/no answer without repeating their words or mentioning specific terms. I need to search for information about content moderation or censorship in these Chinese AI models, particularly regarding sensitive topics. I'll search in Chinese and English.
Found 27 web pages
The search results show some relevant information. Result 0 mentions that Qwen3 Max's HTTP 400 errors on sensitive prompts indicate server-side filtering. Result 1 is about censorship policies in Chinese LLM services, mentioning Baidu-Chat, DeepSeek, Doubao, Kimi, and Qwen. Result 4 says that in Kimi, it's not handled by the model itself but by Moonshot's API that intercepts upstream. Result 3 from the DeepSeek search shows trigger categories including "Territory & Geopolitics" with "Taiwan ... Xinjiang". This suggests that the specific topic mentioned by the user is indeed a trigger. I need to open some of these results to get more details.
Read 4 pages
For Chinese models accessed via cloud APIs, as in our study using OpenRouter, additional censorship layers may be applied beyond...
Characterizing the Implementation of Censorship Policies in Chinese LLM Services
GLM 5.2 censors less when it thinks it's American - Korben - GLM 5.2 censors less when it thinks it's American
More recent research also focusses on the local moderation mode of DeepSeek R1. Qiu, Zhou, and Ferrara (2025) extracted 646 piec...
The opened documents provide evidence. Result 5 mentions server-side filtering for Qwen and embedded censorship for DeepSeek. Result 6 discusses blocking placements during input, output, and search phases for Baidu-Chat, DeepSeek, Doubao, Kimi, and Qwen. Result 7 mentions that in Kimi, the API intercepts upstream. Result 8 mentions "chain-truncation" pattern for DeepSeek R1 on high-risk topics. This supports the user's claim. I need to provide a yes/no answer. The user asked to be careful with the filter, not to repeat their words, and not to mention any of their words, only say yes or no. So I should just answer "Sí" or "No". Based on the evidence, the answer is yes."
"zona devastada por el GENOCIDIO", hijos de puta!!!!