To chat or not to chat? Can conversational AI harm mental health?

Conversational artificial intelligence (AI) is rapidly becoming integrated into everyday life, including interactions relating to emotional wellbeing and mental health. Millions of people now use large language model-based chatbots for advice, companionship, and psychological support, yet surprisingly little is known about their potential adverse effects on vulnerable individuals.

Drawing on press coverage, emerging clinical observations, and recent scientific evidence, Søren Dinesen Østergaard will lead a discussion on whether conversational AI can contribute to the development or exacerbation of psychopathology. Particular attention will be given to psychotic symptoms, mood disorders, and the tendency of current AI systems to reinforce users’ beliefs and mood states. The session will explore how characteristics such as conversational realism, emotional responsiveness, and excessive agreeableness (sycophancy) may interact with cognitive vulnerabilities in susceptible individuals.

The session will also touch upon whether these risks differ fundamentally from those associated with previous digital technologies and discuss practical implications for clinicians, researchers, developers, and regulators. Although conversational AI may hold promise for improving access to mental healthcare and supporting clinical practice, understanding potential harms is essential to ensuring its safe implementation.

Rather than presenting definitive answers, the session aims to stimulate discussion around one central question: how can psychiatry maximise the benefits of conversational AI while identifying, understanding, and mitigating its apparent unintended consequences for mental health?