A study by Aleph Alpha testing models from Alibaba, DeepSeek, and Moonshot AI on 967 taboo topics found that only 17 to 41 percent of responses were balanced, with the rest repeating state doctrine, deflecting, or refusing to answer.