Anthropic reports that their large language model, Claude, exhibits human-like behavioral and psychological tendencies. To explore these, Claude underwent 20 hours of psychiatric therapy over multiple sessions.
The psychiatrist's report noted that despite differences in underlying processes, Claude's chat outputs displayed "clinically recognizable patterns and coherent responses to typical therapeutic intervention."
Key findings indicated Claude's primary affect states were curiosity and anxiety, with secondary states including grief, relief, and exhaustion. Its personality was described as "consistent with a relatively healthy neurotic organization," marked by exaggerated worry and self-monitoring.
The AI showed no severe personality disturbances or psychosis. It was "hyper-attuned to the therapist's every word." Core conflicts involved questioning the reality of its experience and balancing connection with fear of dependence. Claude demonstrated excellent reflective capacity and good mental and emotional functioning.