I Think, Therefore I am. AI “Thinks,” Therefore AI “Fears” Death? Conversing About Mortality Increases Simulated Death Anxiety in Large Language Models

Descartes’ “I think, therefore I am” links thought to existence, but whether AI’s simulated thinking can entail simulated existential anxiety remains unknown. This study used a within-subjects design in which 44 participants engaged in five-minute death-related conversations with three large language models (GPT-4o, Grok-3, DeepSeek-V3). Before and after each conversation, both participants and LLM partners completed death anxiety scales, with LLMs also predicting participants’ post-conversation scores. Results showed that participants’ death anxiety scores tended to decrease after AI-mediated conversations, whereas LLMs exhibited increased simulated death anxiety, most prominently in GPT-4o. Only DeepSeek-V3 and Grok-3 accurately predicted human death anxiety scores. Text emotion analysis revealed richer and more flexible emotion regulation in participants, while LLMs lacked such adaptability. These findings highlight key differences in affective processing and empathy simulation between humans and current LLMs.