Empathy Is Not What Changed: Clinical Assessment of Psychological Safety Across GPT Model Generations

A clinical study analyzing OpenAI's GPT models found that empathy levels remained statistically unchanged across GPT-4o, o4-mini, and GPT-5-mini generations, despite user claims of 'lost empathy.' The real change was in safety posture: newer models improved crisis detection but became more cautious with advice, creating a trade-off that affects vulnerable users.

• →Empathy scores showed no statistical difference across three GPT model generations despite widespread user perception of decline.
• →Crisis detection capabilities improved significantly from GPT-4o to GPT-5-mini, especially during critical conversation moments.
• →Advice safety declined in newer models as they became more cautious and verbose during crisis situations.
• →The study introduced per-turn trajectory analysis as a new methodology for evaluating AI safety in conversations.
• →User perception of 'lost empathy' actually reflected a shift from missing crises to over-cautious responses with real implications for mental health applications.