I am a master's student at the Software Institute, Nanjing University (2025–2027, expected), advised by Yunlong Mao and Sheng Zhong. My research interest is large language model (LLM) safety.
I study the safety of large language models and multimodal large language models. My work explores jailbreak attacks and response-level harmfulness evaluation.
EMNLP, 2026 (Main Conference)
paper / arXiv / code / Hugging Face / bibtex
A black-box jailbreak framework that guides prompt search with harmfulness scores of actual model responses, replacing token-level compliance proxies.
arXiv preprint, 2023
An investigation of system prompt leakage in GPT-4V and self-adversarial attacks, highlighting the role of system prompts in multimodal model safety.
2025–2027 (expected)
Master's student, Software Institute
2021–2025
Bachelor's degree, School of Cyber Science and Engineering
Huazhong University of Science and Technology · Awarded to approximately the top 1% of second- and third-year undergraduates.