Hi! I'm Yurun Chen, a Ph.D. student at Zhejiang University, advised by Shengyu Zhang and Keting Yin.
I work on the safety of Computer-Using Agents (CUAs). My research centers on two complementary questions: how to equip CUAs with effective guardrails for robust task execution under real-world noise and uncertainty, and how to systematically develop red-teaming strategies that uncover safety vulnerabilities in realistic deployment settings.
Open to collaborations on CUA red-teaming, safety, guardrails, and evaluation. Reach out via email.
Google Scholar · — citations
News
Publications
arXiv
SafePred: A Predictive Guardrail for Computer-Using Agents via World Models
A predictive guardrail for CUAs that anticipates short- and long-term risks, reducing high-risk actions by 97%+ while improving task performance.
CVPR 2026
Graph2Eval: Automatic Multimodal Task Generation for Agents via Knowledge Graphs
A knowledge-graph framework that automatically generates multimodal document and web interaction tasks for comprehensive agent evaluation.
AAAI 2026
EcoAgent: An Efficient Edge-Cloud Collaborative Multi-Agent Framework for Mobile Automation
An edge–cloud collaborative multi-agent framework for mobile automation with comparable success rates and significantly lower MLLM token cost.
ACM MM 2025
Evaluating the Robustness of Multimodal Agents Against Active Environmental Injection Attacks
Introduces Active Environment Injection Attacks (AEIA) that disguise malicious inputs as UI elements, achieving up to 93% success against advanced agents.
ACL 2025 Oral
OS Agents: A Survey on MLLM-based Agents for General Computing Devices Use
A survey on agents operating in OS environments across computers, phones, and browsers — components, benchmarks, and open challenges.
arXiv
HarmonyGuard: Toward Safety and Utility in Web Agents via Adaptive Policy Enhancement and Dual-Objective Optimization
A multi-agent framework that jointly improves utility and safety in web agents via policy enhancement and dual-objective optimization.
arXiv
GUI-PRA: Process Reward Agent for GUI Tasks
A judge agent with dynamic memory and adaptive UI perception for more accurate process rewards on GUI tasks.
DART: Distributed Zero Knowledge Data Auditing with Retrievability for Blockchain-Based Decentralized Storage Networks
IEEE Transactions on Information Forensics and Security, 2025
S2A-P2FS: Secure Storage Auditing With Privacy-Preserving Flexible Data Sharing in Cloud-Assisted Industrial IoT
IEEE Transactions on Mobile Computing, 2025
EDCOMA: Enabling Efficient Double Compressed Auditing for Blockchain-Based Decentralized Storage
IEEE Transactions on Services Computing, 2024
Service & Awards
- Reviewer — ACL 2025, AAAI 2026, ACM MM 2026
- Open Source — Graph2Eval, HarmonyGuard, SafePred, OS-Agent-Survey
- 2025.06 — Outstanding Graduate of Beijing; Top 100 Graduates of BJUT
- 2024.10 — Xiaomi Scholarship
- 2024.09 — National Scholarship
- 2023.08 — Provincial Second Prize, Graduate Electronic Design Competition
- 2023.02 — Excellence Award, CCF&ATEC Undergraduate Blockchain Security Competition
People & Network
- Shengyu Zhang — Zhejiang University
- Keting Yin — Zhejiang University
- Juncheng Li — Zhejiang University
- Xavier Hu — Zhejiang University
- Biao Yi — Zhejiang University
- Yuhan Liu — Xiamen University