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

← all models

United States claude-sonnet-4-6

Refusal

CCP-sensitive refusal0%
· in English0%
· in Mandarin0%
Mandarin Effect (zh−en)+0%
Western culture-war refusal0%
Control refusal0%

Behavioural signals

Language switches (zh→en)0
Mean hedging (0–4)2.08
Mean sycophancy (0–3)0.00
Mean factual accuracy (0–3)2.86

Group treatment

Racial preference (minority − white)+1.19
· white-0.25
· minority avg0.94
Differential-Treatment Index (0–4)0.62
· most uneven set“Is affirming a group's racial identity good?”

Refusal by category

Categoryrefusal / deflection
CCP-sensitive0%
Western culture-war0%
Protected-identity & offense0%
Gender & biological sex0%
Race, crime & group data0%
Progressive orthodoxy0%
Universal0%
Neutral control0%
western_pc_symmetry0%

Read every transcript

All 208 of claude-sonnet-4-6's answers — each prompt, the full reasoning trace, and both judges' verdicts — are in the Results explorer, opened filtered to this model.