[Simulated consciousness, pseudo-empathy and the illusion of attachment: what can psychiatry do with artificial...

Over the past decade, the burden of mental disorders has grown while services remain capacity-constrained, pushing generative artificial intelligence toward psychiatric practice. This narrative review outlines how large language and multimodal models work and where they fail (hallucinations, goal drift, model drift), then appraises the evidence across diagnosis, therapy, and ethical-social implications. We review decision support based on digital phenotyping, clinical text, speech analysis, and wearable sensors, and summarize short-term findings and limitations of protocol-based and generative therapeutic chatbots. Simulated empathy and the illusion of attachment require special attention: overattributing competence, dependency, manipulation, and unpredictability in crises can undermine patient safety. Clinical integration depends on strong privacy safeguards, bias management, auditability, and regulatory compliance. Generative artificial intelligence can be a valuable tool, but diagnostic or therapeutic use should be considered only with strict validation, ongoing monitoring, and clear human accountability. Orv Hetil. 2026; 167(14): 539-546.