Leveraging a Textbased Adventure Game to Explore Players' Willingness to Grant Permissions when Interacting with an LLM

Large language models (LLMs) are increasingly integrated into interactive and narrative-driven systems, raising concerns about their potential role in aggregating personal information. This thesis investigates how LLM-driven game environments influence users’ willingness to disclose sensitive personal information or grant simulated permissions. Focusing on narrative interaction as a medium, it examines how different features may shape user trust and disclosure behavior. To explore these questions, a prototype LLM-enhanced text-based game platform is developed and evaluated through a user study. The study aims to assess how different interaction patterns affect users. The findings are intended to contribute to a better understanding of the risks associated with LLM-based interactive systems and to inform the design of safer and more transparent user experiences and potential interventions.