Large Language Models Make Mobile Permission Decisions
Researchers ran an online study with more than 300 participants and collected over 14,000 Android permission decisions to test whether LLMs can auto-respond to access prompts. They compared generic and personalized models from multiple providers, finding generic models matched majority choices 70–86% while personalization and explanations shifted alignment but sometimes reduced security when user statements mismatched behavior. The results highlight design risks and need for safeguards in deployment.
Key Points
- 1Collected over 14,000 Android permission choices from 300+ participants, testing generic versus personalized LLMs.
- 2Found generic LLMs matched user majority 70–86%, improving cautious denials for sensitive permissions.
- 3Shows personalization and explanations can change alignment and security, requiring design safeguards and verification.
Scoring Rationale
Provides empirical, actionable insights on LLM-based permission decisions but limited by experimental setting and threat-model constraints.
Sources
Public references used for this report.
Practice with real Logistics & Shipping data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Logistics & Shipping problems
