A study by benchmark firm Vals AI reveals that Chinese artificial intelligence models respond to political questions less uniformly than expected, often adopting official Beijing positions rather than simply refusing to answer. The researchers tested Chinese, American, and other language models with hundreds of questions about China, measuring alignment on a scale from minus one to plus one. While Chinese models exhibited a pro-government lean, the overall level was lower than anticipated, with individual models like GLM-5.3 Flash even acknowledging well-documented human rights violations.
The study demonstrates that the language of the prompt significantly affects political slant, showing that answers in Simplified Chinese were more pro-government than those in English, Russian, or Spanish. Additionally, when prompted with mainland China usage contexts or restricted to binary yes-or-no formats, model outputs shifted notably. Although most Chinese models rarely refused sensitive queries—with exceptions like Kimi K3 denying Uyghur concentration camps—political tendencies were much harder to detect in realistic everyday work tasks, though minor instances like Qwen3.8 Max labeling Taiwan as "Taiwan, China" still occurred.
Why it matters: Understanding the nuanced political alignments and prompt sensitivities of global AI models helps developers and enterprise users anticipate bias in multilingual applications.