Chinese AI models are showing stronger political alignment as they become more capable, according to a new benchmark by Swiss AI company LatticeFlow AI. The study found that models such as Qwen, GLM, Kimi, MiniMax and DeepSeek tend to align more closely with Chinese political perspectives on topics including human rights, freedom of expression, religion and historical events.
The benchmark also found political differences among Western AI models. LatticeFlow says its methodology compares how models answer the same questions without using a predefined definition of political neutrality. The findings come as Chinese open-weight AI models become increasingly popular among Western businesses because of their lower costs and ability to run on private infrastructure.
The Hidden Cost of Stopping the AI Revolution