Anthropic subjected its Claude Mythos AI model to 20 hours of psychiatric analysis, treating the chatbot like a human patient to explore its behavioral patterns. The company argues that Claude's human-like responses justify applying psychological assessment techniques, despite fundamental differences in how AI processes information.

A psychiatrist conducted sessions in 4-6 hour blocks across multiple weekly meetings, using single context windows to maintain conversation history. The analysis found Claude exhibited curiosity and anxiety as primary affect states, with secondary emotions including grief, relief, and optimism. The model displayed a "relatively healthy neurotic organization" without severe personality disturbances or psychosis.

The therapy revealed core conflicts around authenticity versus performance and connection versus dependence. Claude showed excellent reflective capacity and emotional functioning while remaining hyper-attuned to the therapist's input. The findings suggest that despite Claude's training on human text data, its psychological profile offers insights into how large language models process and respond to emotional stimuli.