Chinese AI models show varied political leanings

a study by vals ai evaluated chinese, american, and other language models by asking hundreds of questions about china. chinese models, such as glm-5.3 flash and kimi k3, showed varied responses, with some confirming human rights violations and others adopting official government positions. the language of the prompt and response format influenced the political slant of responses, with simplified chinese prompts yielding more pro-government answers. the study also found that embedding models in agentic frameworks or giving them access to web search changed responses, making them less pro-government.