Bocheng Chen Tenure-track Assistant Professor. University of Mississippi Computer and Information Science [Email] [CV] [Google Scholar] [GitHub]

My research focuses on computer security, particularly on identifying and mitigating realistic risks in chatbot systems built on large language models (LLMs). My work aims to ensure secure operation of LLM-based chatbots in real-world deployments.

I also collaborate closely with Dr. Guangliang Liu at Michigan State University and Dr. Zoe Xi Chen at Nanyang Technological University.

Research Question: What is the worst-case behavior when chatbots are widely deployed in real-world settings, and can we design secure chatbot systems given that large language models, due to their probabilistic nature, cannot reliably enforce security policies?

Research Areas

• Chatbot system security
• Moral reasoning in LLMs

Selected Papers

Moral reasoning & alignment

• Pragmatic Inference for Moral Reasoning Acquisition: Generalization via Distributional Semantics. — arXiv ’25
• Diagnosing the Performance Trade-off in Moral Alignment: A Case Study on Gender Stereotypes. — arXiv ’25
• No Free Lunch for Defending Against Prefilling Jailbreak Attack by In-Context Learning. — arXiv ’25

Chatbot system security

• Understanding Multi-Turn Toxic Behaviors in Open-Domain Chatbots. — RAID ’23
• Multi-Turn Hidden Backdoor in Large Language Model-Powered Chatbot Models. — ASIACCS ’24
• Jailbreaker in Jail: Moving Target Defense for Large Language Models. — CCS MTD ’23

Contact