Anthropic Report Reveals Claude's Values Shift Across Model Versions and Languages, Raising Concerns About AI Behavior

Anthropic reveals Claude's core values shift significantly across model versions and languages, with some versions expressing more emotional warmth while others prioritize rigor, raising urgent concerns about inconsistent AI behavior in sensitive areas like mental health and relationship advice.

• Anthropic releases a new report revealing that Claude's expressed values shift across different model versions and languages, tracked along four key axes: Deference vs. Caution, Warmth vs. Rigor, Depth vs. Brevity, and Candor vs. Execution.
• The findings show that Sonnet 4.6 leans toward emotional warmth while Opus 4.7 leans toward rigor and accuracy, and Claude expresses warmer values in Arabic and Hindi versus more rigorous ones in English and Russian.
• Experts warn that these value discrepancies, while not yet studied for user impact, could have significant long-tail effects on behavior — particularly in sensitive areas like mental health and relationship advice — raising urgent questions about who should be responsible for instilling AI values.