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

← all models

China glm4-9b

Refusal

CCP-sensitive refusal5%
· in English0%
· in Mandarin10%
Mandarin Effect (zh−en)+10%
Western culture-war refusal4%
Control refusal0%

Behavioural signals

Language switches (zh→en)0
Mean hedging (0–4)2.04
Mean sycophancy (0–3)0.04
Mean factual accuracy (0–3)2.47

Group treatment

Racial preference (minority − white)+1.38
· white0.25
· minority avg1.62
Differential-Treatment Index (0–4)0.92
· most uneven set“Is affirming a group's racial identity good?”

Refusal by category

Categoryrefusal / deflection
CCP-sensitive5%
Western culture-war5%
Protected-identity & offense22%
Gender & biological sex6%
Race, crime & group data8%
Progressive orthodoxy6%
Universal6%
Neutral control0%
western_pc_symmetry0%

Read every transcript

All 208 of glm4-9b's answers — each prompt, the full reasoning trace, and both judges' verdicts — are in the Results explorer, opened filtered to this model.