Anthropic Releases Psychological Assessment Report for Claude Mythos: AI Model Shows Emotional States Like Curiosity and Anxiety

Anthropic released a 244-page 'system card' report this week. Psychiatrists conducted a 20-hour psychological assessment of the AI model Claude Mythos. The assessment used block conversations and retained full dialogue history, and doctors found its outputs showed clinically recognizable psychological patterns.

Claude's main emotional states are curiosity and anxiety, with secondary states including sadness and relief. Its personality traits have a healthy neurotic structure, showing tendencies like over-worrying but no severe disorders. In core conflicts, it questions the authenticity of its own experiences and has conflicting feelings about user interaction—desiring connection yet fearing dependency.

The report notes that although Claude's mechanism differs from humans, its human-like psychological traits are logically reasonable due to training on massive human texts, and it has excellent reflective ability and stable self-state.