LLMs for Personalized Access Control Decisions

A research paper investigates whether large language models (LLMs) can make personalized access control decisions to reduce user cognitive burden. The study focuses on smartphone app permission requests, using a dataset of 307 user privacy statements and 14,682 permission decisions. The approach involves a lightweight setup phase where users express security preferences in natural language, enabling LLMs to make dynamic, context-aware decisions aligned with those preferences. This aims to address the increasing complexity and automation of systems that often overwhelm users, leading to suboptimal choices. The paper is published on arXiv under ID 2511.20284.