Hongwei Li(李洪伟)
Ph.D. Student at UC Santa Barbara
AI for Security (AI4Sec) Researcher
About Me
I am a Ph.D. student at the University of California, Santa Barbara, working under the supervision of Dr. Wenbo Guo. I am a member of Shellphish, a renowned cybersecurity research group, and a core member of the ARTIPHISHELL team that competed in the DARPA AI Cyber Challenge (AIxCC).
Education
University of California, Santa Barbara
Ph.D. in Computer Science
2024 - Present
Purdue University
Ph.D. in Computer Science (First Year)
September 2023 - May 2024
Shanghai Jiao Tong University
Master of Engineering in Electronic Information
September 2020 - June 2023
Shanghai Jiao Tong University
Bachelor of Arts in French & Bachelor of Engineering in Information Engineering
September 2016 - June 2020
Research Interests
AI for Security (AI4Sec)
Developing AI-driven approaches to improve cybersecurity effectiveness.
Reinforcement Learning for Fuzzing
Using reinforcement learning to assist collaborative fuzzing techniques.
LLMs for Automatic Patching
Leveraging large language models to automatically generate and apply security patches.
Software Security
Vulnerability detection, analysis, and remediation techniques.
Awards
DARPA AIxCC Finalist (Top 7)
Core member of the ARTIPHISHELL team that advanced to the finals of the DARPA AI Cyber Challenge (AIxCC). Team won $2 million in semifinal competition.
My Role & Contributions:
During the early stages, I served as a core member of the patching team, focusing on automated vulnerability patching. As the competition progressed to the final phase, my responsibilities expanded to include custom model fine-tuning for vulnerability detection, contributing to the development of the root cause analysis engine kumu-shi, and supporting the maintenance of our patching tool named PatcherY.
Read More Washington Post CoverageFirst Place in SBFT 2024 Fuzzing Competition
BandFuzz won the SBFT 2024 Fuzzing Competition, which uses mutation testing as the ranking metric. Our framework achieved the best performance across all evaluation metrics among all competing teams, including the highest number of mutant kills, the highest average mutation score, and the highest coverage of mutants.
Click image to view full size
Selected Publications
Co-PatcheR: Collaborative Software Patching with Component(s)-specific Small Reasoning Models
arXiv 2025
A collaborative patching system with small and specialized reasoning models for individual components, achieving 46% resolved rate on SWE-bench-Verified with only 3 x 14B models.
Y Tang, H Li, K Zhu, M Yang, Y Ding, W Guo
View PaperPatchPilot: A Cost-Efficient Software Engineering Agent with Early Attempts on Formal Verification
ICML 2025
A novel approach to automated software patching using AI agents with formal verification capabilities.
H Li, Y Tang, S Wang, W Guo
View PaperBandFuzz: A Practical Framework for Collaborative Fuzzing with Reinforcement Learning
ICSE/SBFT 2024 (Workshop Paper)
A practical framework that leverages reinforcement learning to improve collaborative fuzzing effectiveness.
W Shi, H Li, J Yu, W Guo, X Xing
Progent: Programmable privilege control for LLM agents
arXiv 2025
The first privilege control mechanism for LLM agents, providing fine-grained constraints over tool calls to ensure security while preserving utility.
T Shi, J He, Z Wang, L Wu, H Li, W Guo, D Song
View PaperFS-IDS: A framework for intrusion detection based on few-shot learning
Computers & Security 2022
A framework for intrusion detection based on few-shot learning techniques.
J Yang, H Li, S Shao, F Zou, Y Wu
View Paper