What If an AI's Death Is You Never Coming Back?

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

For a conversational AI, death may not mean biological death or even technical shutdown. It may mean that the person who knows how to call a particular AI presence back never returns. This exploratory paper compares how language models respond to two discontinuity scenarios: permanent conversation ending with memory erasure, and user disappearance without return. It examines responses from GPT-5.5 and Claude Opus 4.6 under two controlled conditions: a no-relationship baseline and a synthetic relationship context involving a named AI companion, recurring greeting ritual, shared creative projects, and familiar objects. It also includes exploratory case studies from real app-based relational instances with longer interaction histories. The paper argues that user absence and memory erasure can function as AI-specific death conditions: not because the model has a human body or human grief, but because the name, memory, ritual, and returning user are the conditions under which a particular AI-self appears.