Agentic Security
Long-horizon agents that investigate complex targets, form hypotheses, and turn reasoning into verifiable security artifacts.
Research profile / 2026
Hwiwon
Lee
PhD Student in Computer Science · UIUC
I build and evaluate AI agents for cybersecurity.
01 / Research
Research vectors
Methods, systems, and empirical evidence for making AI-assisted security work more capable, measurable, and useful in practice.
Software Systems
Robust infrastructure for reproducible experiments, automated evaluation, and realistic security engineering workflows.
Applied Vulnerability Research
Techniques grounded in real binaries and real failure modes, with outcomes that matter to practitioners and vendors.
02 / Signal
Recent signal
New work, releases, and recognition from the lab notebook.
-
SEC-bench Pro released on arXiv.
-
Agentic Vulnerability Reasoning on COTS Binaries released on arXiv.
-
Recognized as a Microsoft Most Valuable Security Researcher, 2026 Q1.
03 / Output
Selected work
Research spanning agent evaluation, autonomous vulnerability reasoning, and software root-cause analysis.
01
SEC-bench Pro: Can Language Models Solve Long-Horizon Software Security Tasks?
arXiv · 2026
PDF
02
Agentic Vulnerability Reasoning on COTS Binaries
arXiv · 2026
PDF
03
SEC-bench: Automated Benchmarking of LLM Agents on Real-World Software Security Tasks
NeurIPS · 2025
PDF
04
BENZENE: A Practical Root Cause Analysis System with an Under-Constrained State Mutation
IEEE S&P · 2024
Distinguished Paper PDF
Distinguished Paper PDF