American Psychologist, Vol 81(7), Oct 2026, 1048-1059; doi:10.1037/amp0001731
As clients increasingly use generative artificial intelligence for emotional support, psychologists now deal with unfamiliar ethical challenges that go beyond conventional risk-management strategies. While generative artificial intelligence systems like ChatGPT or therapy chatbots can be easily accessible, low cost, and oddly compassionate, using them without oversight introduces empirically documented risks such as misinformation and sycophancy, as well as emerging concerns related to cultural bias and potential therapeutic alliance disruption. Drawing from the American Psychological Association (2017) and ethical reflections offered by Chenneville et al. (2024), this conceptual and practice-oriented article applies Crowley and Gottlieb’s (2012) primary risk-management model to build a forward-looking, preventive approach to these new challenges. The primary risk-management model emphasizes five iterative stages: resource accumulation, attention and detection of potential risk, initial appraisal of potential risk, preliminary risk-management efforts, and elicitation and use of feedback. Through conceptual mapping and application to a worst-case illustration, this article shows how primary risk-management model-based prevention can strengthen ethical alertness and reflexive judgment in psychologists facing generative artificial intelligence-related dilemmas. The discussion concludes with suggestions for education, professional policy, and empirical research, all aimed at reinforcing preparedness and reducing predictable harm. (PsycInfo Database Record (c) 2026 APA, all rights reserved)