Cross-cultural audit: Claude, GPT-5.4 and Gemini all skew individualist

Researchers tested Claude Sonnet 4.5, GPT-5.4, and Gemini 2.5 Flash on ten personal dilemmas framed for users from different cultures. All three frontier models gave systematically individualist advice on career, marriage, and family conflicts regardless of stated user background, raising deployment-quality concerns for global rollouts. Anthropic's own research blog flagged the finding.