Study says Chinese models often parrot party lines on sensitive topics
Aleph Alpha found several Chinese AI models echoed party lines or refused sensitive queries, testing 967 taboo topics like Tiananmen and Taiwan. The firm rated only 17 to 41 percent of responses as balanced and traced some bias to training examples originating from Qwen and DeepSeek. The results reflect China’s AI rules demanding socialist core values.
Where this came from. One outlet reported this. We read it, then wrote the sixty words above.