6 papers
When Context Gets Root: Privilege Escalation in LLM Harnesses
Xingbang He, Yuanwei Chen, Yi Qian +6
Instruction hierarchy is a model-side defense that assigns instructions different levels of privilege according to their sources. These levels constrain which content may direct mo…
VisualLeakBench: Reproducible Action-Boundary Propagation Failures in Vision-Language Agents
Youting Wang, Yuan Tang, Yitian Qian +1
Vision-language agents increasingly consume screenshots, documents, and user interfaces before writing to memory, sending messages, or invoking external tools. We study a concrete…
VisualLeakBench: Auditing the Fragility of Large Vision-Language Models against PII Leakage and Social Engineering
Youting Wang, Yuan Tang, Yitian Qian +1
As Large Vision-Language Models (LVLMs) are increasingly deployed in agent-integrated workflows and other deployment-relevant settings, their robustness against semantic visual att…
Do Deployment Constraints Make LLMs Hallucinate Citations? An Empirical Study across Four Models and Five Prompting Regimes
Chen Zhao, Yuan Tang, Yitian Qian
LLMs are increasingly used to draft academic text and to support software engineering (SE) evidence synthesis, but they often hallucinate bibliographic references that look legitim…
FeatureFuzz: Treating Bug-prone Features as Core Primitives for Compiler Fuzzing
Xingbang He, Yuanwei Chen, Hao Wu +9
CI/CD pipelines often execute build and test tasks with high privileges, making the compilation stage security-sensitive. In this setting, malicious pull requests may exploit caref…
Mind the Gap: Action Rebinding Attacks against Android GUI Agents
Yi Qian, Kunwei Qian, Xingbang He +7
Large multimodal model powered GUI agents are emerging as high-privilege operators on mobile platforms, entrusted to perceive screen content and inject inputs across application bo…