Paper 2510.24411
OS-Sentinel: Towards Safety-Enhanced Mobile GUI Agents via Hybrid Validation in Realistic Workflows
- Published
- Oct 2025
- Research lab
- Independent
- Citations
- 15
- GitHub
- 49 stars
01 In brief
Summary
This paper introduces OS-Sentinel, a hybrid framework for detecting safety risks in mobile GUI agents powered by Vision-Language Models (VLMs).
The authors first construct MobileRisk-Live, a dynamic Android emulator sandbox that records GUI observations, agent actions, and a System State Trace (T_sys) capturing underlying system metadata.
From this, they derive MobileRisk, a benchmark of 204 human-annotated trajectories (102 unsafe, 102 safe) with fine-grained labels including trajectory-level binary safety, the first unsafe step, and a ten-category risk taxonomy.
OS-Sentinel combines a Formal Verifier that uses deterministic checks (system state integrity hashing, sensitive keyword detection, and pattern matching) with a VLM-based Contextual Judge that assesses semantic risks from screenshots and actions.
Experiments show OS-Sentinel outperforms rule-based and VLM/LLM-as-a-Judge baselines by 10-30% in accuracy and F1 across multiple backbones (GPT-4o, Claude, Qwen2.5-VL) at both step and trajectory levels.
The framework supports strict and consensus modes, and analysis shows the hybrid approach provides broader and more balanced detection across risk categories.
The authors also validate that frozen trajectories closely approximate live environment performance, supporting reproducible safety research.
02 From the paper
Abstract
Computer-using agents powered by Vision-Language Models (VLMs) have demonstrated human-like capabilities in operating digital environments like mobile platforms. While these agents hold great promise for advancing digital automation, their potential for unsafe operations, such as system compromise and privacy leakage, is raising significant concerns. Detecting these safety concerns across the vast and complex operational space of mobile environments presents a formidable challenge that remains critically underexplored. To establish a foundation for mobile agent safety research, we introduce MobileRisk-Live, a dynamic sandbox environment accompanied by a safety detection benchmark comprising realistic trajectories with fine-grained annotations. Built upon this, we propose OS-Sentinel, a novel hybrid safety detection framework that synergistically combines a Formal Verifier for detecting explicit system-level violations with a VLM-based Contextual Judge for assessing contextual risks and agent actions. Experiments show that OS-Sentinel achieves 10%-30% improvements over existing approaches across multiple metrics. Further analysis provides critical insights that foster the development of safer and more reliable autonomous mobile agents. Our code and data are available at https://qiushisun.github.io/OS-Sentinel-Home/.