Security-Fidelity Tradeoffs: The Hidden Cost of Prompt Injection Defense
SecFid reveals that prompt-injection defenses can gain security by suppressing untrusted text, trading away fidelity on tasks that must preserve it.
PhD student in Computer Science, Siebel School of Computing and Data Science at UIUC
I study how to make AI systems secure and reliable.
Evaluation, language model security, and deployment reliability.
✦ ICML 2026 Spotlight Sloan Scholar SURGE Fellow
Selected work
Highlighted research
- ↗
- ↗
That's Deprecated! Understanding, Detecting, and Steering Knowledge Conflicts in Language Models for Code Generation
A study of how language models handle conflicts between learned programming knowledge and prompt-provided updates, with probes and activation steering for code generation.
About
Research and background
I am a PhD student in Computer Science at the Siebel School of Computing and Data Science, advised by Professor Haohan Wang.
My current work studies language model security and evaluation. SecFid examines the tradeoff between security and fidelity in prompt-injection defenses. That's Deprecated studies how language models respond when prompts conflict with learned programming knowledge.
Before Illinois, I studied computer science, statistics, and mathematics at the University of Iowa.
- Education
- PhD in Computer Science, UIUCBS Computer Science, BS Statistics, BA Mathematics, Iowa
- Advisor
- Professor Haohan Wang
- Recognition
- ICML 2026 Spotlight Sloan Scholar, UIUC SURGE Fellow Earlier: Emerging Technologies Scholarship and Collins Aerospace AI/ML Scholarship
- Teaching and leadership
- Excellent Teaching List, STAT 107 Fall 2024, Spring 2025 President, ACM at the University of Iowa, 2021-2023