Changyue Jiang (蒋昌跃)
I am Changyue Jiang (蒋昌跃), a Ph.D. student at Fudan University (复旦大学) and Shanghai Innovation Institute (SII, 上海创智学院), advised by Prof. Xudong Pan (潘旭东) and Prof. Min Yang (杨珉).
My research focuses on AI security and safety for agentic systems. I study how mobile agents, computer-use agents, and retrieval-augmented LLM applications can fail, and build policy-aware auditing, thought and trajectory correction, and agent-protection tools that make their behavior more trustworthy, controllable, and auditable.
Email: cyjiang24@m.fudan.edu.cn
Research Interests
- AI Security
- AI Safety
- Agent Safety
- RAG Security
我是蒋昌跃，复旦大学与上海创智学院全时培养在读博士生，导师为潘旭东和杨珉老师。
我的研究关注智能体系统中的 AI 安全。我围绕规则感知的智能体轨迹安全审计、智能体推理和轨迹纠错、端侧智能体全流程安全防护开展研究，希望提升智能体行为的安全性、可控性与可审计性。
邮箱：cyjiang24@m.fudan.edu.cn
研究方向
- AI 安全
- AI Safety
- 智能体安全
- RAG 安全
Publications论文
Think Twice Before You Act: Enhancing Agent Behavioral Safety with Thought Correction
ICML 2026 2026
This work improves behavioral safety for LLM agents by correcting risky thoughts before action execution.
MATE: Policy-Aware Security Auditing for Mobile Agents via Synthesis-Driven Trajectory Learning
USENIX Security 2026 2026
MATE is a policy-aware security auditing framework for mobile agents, using synthesis-driven trajectory learning to find unsafe behaviors.
MirrorGuard: Toward Secure Computer-Use Agents via Simulation-to-Real Reasoning Correction
ACM CCS 2026 2026
MirrorGuard improves the security of computer-use agents by using simulation-to-real reasoning correction.
Feedback-Guided Extraction of Knowledge Base from Retrieval-Augmented LLM Applications
arXiv preprint 2024
This work studies knowledge-base extraction attacks against retrieval-augmented LLM applications and introduces a feedback-guided extraction method.
See the Publications page for publication details.
更多论文信息见 Publications 页面。
Open Source Projects开源项目
Qise
A local-first security layer for AI agents. Qise helps users protect agents such as Codex, OpenClaw, Claude Code, and custom agents by scanning risky integrations, routing model traffic through a local guard proxy, blocking dangerous actions, and recording readable local security events. 面向 AI 智能体的本地优先安全层。Qise 通过扫描高风险集成、将模型流量路由到本地防护代理、阻断危险动作并记录可读的本地安全事件，帮助用户保护 Codex、OpenClaw、Claude Code 和自定义智能体。
Snowl
A framework-agnostic evaluation engine for AI agents. Snowl helps run agents against built-in or custom benchmarks, produce reproducible scores, compare models fairly, and record full evaluation artifacts for safety and capability analysis. 面向 AI 智能体的框架无关评测引擎。Snowl 支持将智能体接入内置或自定义基准，生成可复现分数，公平比较不同模型，并记录完整评测产物，用于安全性和能力分析。
snowl-mobile
A mobile-agent evaluation and safety research toolkit for running, reproducing, and analyzing mobile-agent tasks on Android devices and emulators, with workflows oriented toward mobile agent safety benchmarks and experiment automation. 面向移动智能体评测与安全研究的工具包，用于在 Android 设备和模拟器上运行、复现和分析移动智能体任务，并支持移动智能体安全基准和实验自动化流程。
Academic Service & Activities
- Invited to attend the Frontier AI Safety Dinner hosted by Safe AI Forum (SAIF), Seoul, 2026.
- Co-authored and released an agent security group standard with the Shanghai Computing Power Network Association.
- Organizer, AI Deception Workshop, Fudan University & SAIF & Concordia AI, 2025.
Internship
- R&D Engineer Intern, ByteDance, Jun. 2023 - Sep. 2023.
Honors & Awards
- Outstanding Achievement Award (Top 6), First Research and Innovation Education Achievement Exchange, Shanghai Innovation Institute, 2026.
- Finalist (Top 6), Innovation Breakthrough Zhenyi Scholarship Defense, Shanghai Innovation Institute, 2025.
- Research Excellence Academic Scholarship, Fudan University, 2025.
- Graduate Academic Excellence Scholarship, Fudan University, 2024.
- Outstanding Graduate Student, Wuhan University, 2023.
- Graduate Academic Excellence Scholarship, Wuhan University, 2023.
- Graduate Excellence Scholarship, Wuhan University, 2023.
- Graduate Excellence Scholarship, Wuhan University, 2022.
- University-level Outstanding Graduate, 2021.
- National Encouragement Scholarship, 2020.
- University-level Merit Student, 2020.
- National Encouragement Scholarship, 2019.
- University-level Merit Student, 2019.
- National First Prize, China Robot Competition, 2019.
- National Second Prize, 5th China Youth Volunteer Service Project Competition, 2020.
- National Third Prize, 13th China Collegiate Computing Design Competition, 2020.
- National Third Prize, 20th China Graduate Mathematical Modeling Competition, 2023.
- Central South Regional First Prize, 13th China Collegiate Computing Design Competition, 2020.
- Central South Regional Second Prize, 13th China Collegiate Computing Design Competition, 2020.
- Central South Regional Second Prize, 12th China Collegiate Computing Design Competition, 2019.
- Hubei Provincial Second Prize, 11th Lanqiao Cup National Software and IT Professional Talent Competition, C/C++ Programming, 2020.
- Hubei Provincial Third Prize, Contemporary Undergraduate Mathematical Contest in Modeling, 2019.
学术服务活动
- 受邀参加 Safe AI Forum (SAIF) 举办的 Frontier AI Safety Dinner，首尔，2026。
- 与上海市算力网络协会联合撰写并发布智能体安全团体标准。
- 组织者，AI Deception Workshop，复旦大学 & SAIF & Concordia AI，2025。
实习经历
- 字节跳动，研发工程师实习生，2023.06 - 2023.09。
竞赛、奖学金等
- 2026 年上海创智学院首届研创学育人成果交流优秀成果奖（前 6 名）。
- 2025 年上海创智学院创新突破臻益奖学金答辩入围（前 6 名）。
- 2025 年复旦大学研究型优秀学业奖学金。
- 2024 年复旦大学研究生优秀学业奖学金。
- 2023 年武汉大学优秀研究生称号。
- 2023 年武汉大学研究生优秀学业奖学金。
- 2023 年武汉大学研究生卓越奖学金。
- 2022 年武汉大学研究生卓越奖学金。
- 2021 年校级优秀毕业生称号。
- 2020 年国家励志奖学金。
- 2020 年校级三好学生称号。
- 2019 年国家励志奖学金。
- 2019 年校级三好学生称号。
- 2019 年中国机器人大赛全国一等奖。
- 2020 年第五届中国青年志愿服务项目大赛全国二等奖。
- 2020 年第十三届中国大学生计算机设计大赛全国三等奖。
- 2023 年第二十届中国研究生数学建模竞赛全国三等奖。
- 2020 年第十三届中国大学生计算机设计大赛中南赛区一等奖。
- 2020 年第十三届中国大学生计算机设计大赛中南赛区二等奖。
- 2019 年第十二届中国大学生计算机设计大赛中南赛区二等奖。
- 2020 年第十一届蓝桥杯全国软件和信息技术专业人才大赛 C/C++ 程序设计湖北省二等奖。
- 2019 年高教社杯全国大学生数学建模竞赛湖北赛区三等奖。