globusjournal

Persistent, cross-session memory is becoming a standard feature of generative AI assistants, allowing systems to recall a user’s name, preferences, and prior disclosures across conversations. This article examines how enabling versus disabling conversational memory affects two outcomes central to human-AI communication: the length of users’ self-disclosures and their perceived safety when sharing personal information. Integrating communication privacy management theory, computers-are-social-actors research, and recent empirical work on long-term memory in large language model (LLM)-driven chatbots (2019–2025), the article reports an illustrative between-subjects experiment (N = 356) constructed to illustrate how such a true experiment could be analyzed, including the operational and ethical requirements of the memory manipulation; these results are not derived from actual human participants. The article defines the outcome constructs, discusses ethical safeguards for sensitive-topic disclosure research, and provides a concrete action plan for a genuine empirical test.

Keywords: AI, Human-AI Communication, Generative AI, Ethical, Empathy, Automation, Safety, Chatbot Interaction.