Chinese AI models parrot state doctrine or refuse to answer on sensitive topics

Study finds Chinese AI models censor sensitive topics and follow state doctrine

Research by Aleph Alpha found that Chinese AI models like Alibaba's Qwen either refuse to answer questions on politically sensitive topics or provide responses aligned with Chinese government positions, with only 17-41% of answers rated as balanced. The models reportedly avoid topics like Tiananmen Square and other subjects deemed sensitive by Chinese authorities.