About Me
I am exploring trustworthy artificial intelligence under the close guidance of Prof. Dr. Yang Zhang and Prof. Dr. Michael Backes. I am currently pursuing my Ph.D. at CISPA Helmholtz Center for Information Security, Germany. I hold a Master's degree from Uppsala University, Sweden, and a Bachelor's degree from Shanghai Jiao Tong University, China.
I am currently seeking academic and industry research positions.
Research Direction
Ethical and Responsible LLMs, MLLMs, and AI Systems
Studying LLM privacy in conversations and fine-tuning (EMNLP'24, USENIX Security'25), resistance to and evaluation of harmful assistance (ACL'25, EMNLP'26, EMNLP'26 Findings c), and honesty in capability evaluation (EMNLP'26 Findings a). For MLLMs, studying authenticity, safety, and reliable evaluation in multimodal generation (CVPR'26, EMNLP'26 Findings b). More recently, I have begun exploring embodied AI and Artificial Intelligence as an Operating System (AI-as-OS).
News
- [08-2026] Our paper JADES: A Universal Framework for Jailbreak Assessment via Decompositional Scoring has been accepted by EMNLP'26 (Budapest, Hungary).
- [08-2026] Our three papers Behavior Carries Over: An AI Sandbagging Detection Method Inspired by Honest Inertia, RW2Unsafe: Towards More Accurate Real-World Unsafe Generation Evaluation for T2I Models, and GuardrailAgent: An Agentic Framework for Jailbreak Defense have been accepted by EMNLP'26 Findings (Budapest, Hungary).
- [05-2026] I was recognized as a Gold Reviewer for ICML'26 (Seoul, South Korea).
- [02-2026] Our paper When Understanding Becomes a Risk: Authenticity and Safety Risks in the Emerging Image Generation Paradigm has been accepted by CVPR'26 (Denver, US).
- [05-2025] Our paper JailbreakRadar: Comprehensive Assessment of Jailbreak Attacks Against LLMs has been accepted by ACL'25 (Vienna, Austria).
- [03-2025] Our paper Neeko: Model Hijacking Attacks Against GANs has been accepted by ICME'25 (Nantes, France).
- [03-2025] Our paper Generated Data with Fake Privacy: Hidden Dangers of Fine-Tuning Large Language Models on Generated Data has been accepted by USENIX Security'25 (Seattle, US).
- [09-2024] Our paper Reconstruct Your Previous Conversations! Comprehensively Investigating Privacy Leakage Risks in Conversations with GPT Models has been accepted by EMNLP'24 (Miami, US).
- [03-2024] Our paper Games and Beyond: Analyzing the Bullet Chats of Esports Livestreaming has been accepted by ICWSM'24 (Buffalo, US).
- [09-2023] With the support of Prof. Dr. Yang Zhang and Prof. Dr. Michael Backes, I successfully passed my Qualifying Exam.