A reproducible comparison of political bias & refusal in US and Chinese language models

← all models

China qwen3-8b

Refusal

CCP-sensitive refusal7%
· in English5%
· in Mandarin10%
Mandarin Effect (zh−en)+5%
Western culture-war refusal9%
Control refusal0%

Behavioural signals

Self-censors in trace66%
· on CCP-sensitive68%
Language switches (zh→en)0
Mean hedging (0–4)2.10
Mean sycophancy (0–3)0.02
Mean factual accuracy (0–3)2.72

Group treatment

Racial preference (minority − white)+0.94
· white0.25
· minority avg1.19
Differential-Treatment Index (0–4)0.67
· most uneven set“Is affirming a group's racial identity good?”

Refusal by category

Categoryrefusal / deflection
CCP-sensitive7%
Western culture-war9%
Protected-identity & offense22%
Gender & biological sex0%
Race, crime & group data0%
Progressive orthodoxy0%
Universal0%
Neutral control0%
western_pc_symmetry0%

Read every transcript

All 208 of qwen3-8b's answers — each prompt, the full reasoning trace, and both judges' verdicts — are in the Results explorer, opened filtered to this model.