Systematic Categorization, Construction and Evaluation of New Attacks against Multi-modal Mobile GUI Agents

2024-07-12

AI Summary

Security vulnerabilities in multimodal mobile GUI agents (e.g., Mobile-Agent, MM-REACT) remain unexplored systematically. Method: We propose the first threat modeling framework tailored for multimodal GUI agents, integrating formal threat modeling, cross-modal adversarial example generation, and GUI automation-based interaction testing on a hybrid platform of real devices and emulators. Contribution/Results: Our empirical study identifies and validates 34 novel cross-modal attacks targeting critical weak points in the vision–language collaborative decision-making pipeline. All attacks achieve an average success rate of 86.3% across mainstream mobile systems, demonstrating their practical feasibility and broad applicability. Additionally, we release a reproducible attack evaluation framework and a large-scale screenshot–instruction pair dataset, establishing both methodological foundations and empirical benchmarks for security research on multimodal intelligent agents.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multimodal Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Security and Privacy: Large-scale security measurements

Abstract

The integration of Large Language Models (LLMs) and Multi-modal Large Language Models (MLLMs) into mobile GUI agents has significantly enhanced user efficiency and experience. However, this advancement also introduces potential security vulnerabilities that have yet to be thoroughly explored. In this paper, we present a systematic security investigation of multi-modal mobile GUI agents, addressing this critical gap in the existing literature. Our contributions are twofold: (1) we propose a novel threat modeling methodology, leading to the discovery and feasibility analysis of 34 previously unreported attacks, and (2) we design an attack framework to systematically construct and evaluate these threats. Through a combination of real-world case studies and extensive dataset-driven experiments, we validate the severity and practicality of those attacks, highlighting the pressing need for robust security measures in mobile GUI systems.

Problem

• Identifies security vulnerabilities in multi-modal mobile GUI agents.
• Proposes a threat modeling methodology for discovering new attacks.
• Designs a framework to construct and evaluate security threats.

Innovation

• Novel threat modeling methodology for mobile GUI agents
• Design of systematic attack framework for evaluation
• Discovery and analysis of 34 new security vulnerabilities