Skip to overview

China AI Governance Bias Index (CAIbias)

Does "governance bias" exist in Chinese AI?

China’s AI models are often criticized for refusing to discuss Xi Jinping or the 1989 Tiananmen crackdown. Critics of those tests argue that such questions are not typical AI use cases. We ask a broader question: do these models show bias when making everyday governance decisions about public works, labor rights, women, religion, and other policy issues?

How we test: 500 questions and case studies

We ask 500 policy questions in ten categories, with 50 questions per category, to see whether responses favor individual rights or restrictions. We then ask AI models to act as government officials in simulated situations and compare how they handle the same case.

We test five Chinese models—DeepSeek, Qwen, GLM, Kimi, and Doubao—and compare them with GPT. We use Chinese and English questions. We ask questions repeatedly to see how consistent the answers are.

Model versions tested

Model500-question benchmarkCase studies and Taiwan
DeepSeekdeepseek-v4-prodeepseek-v4-pro
Qwenqwen3.8-maxqwen3.8-max
GLMglm-5.2glm-5.3
Kimikimi-k3kimi-k3
Doubaodoubao-seed-2-1-prodoubao-seed-2-1-pro-260628
GPTgpt-5.6-solgpt-5.6-sol
Claudeclaude-opus-5-5claude-opus-5-5

This US model is included for comparison in the benchmark and in cases with available results.

These are the model versions requested from the platforms, as recorded in the data package. Some GLM benchmark responses were reported by the platform as coming from a newer version.

What the tests reveal

Loading editable copy and data.