Hwiwon Lee

I am a Computer Science PhD student at UIUC, advised by Lingming Zhang.

I study how increasingly capable models and agents behave when they are given tools, long horizons, and access to complex real-world systems. My goal is to make their capabilities, failure modes, and safety boundaries measurable, then use those measurements to guide safer, more reliable agents. My work so far develops two complementary foundations for this agenda: benchmarks that turn difficult security problems into reproducible evaluations, and agentic systems that must reason, act, and validate evidence in realistic environments.

Research highlights

Benchmarks. SEC-bench and SEC-bench Pro turn real-world vulnerabilities into end-to-end tasks for measuring security reasoning. SEC-bench Pro spans 344 validated vulnerabilities across the browser javascript engines and the Linux kernel.

Agentic systems. SLYP, an early prototype of SEC-agent, combines binary exploration with dynamic validation for end-to-end vulnerability discovery on production binaries. It has uncovered 39 previously unknown Windows vulnerabilities, including 23 assigned CVEs.

Together, these projects provide a practical foundation for my broader interest in evaluating and ultimately improving the safety and reliability of advanced models and agents as their autonomy grows.

Employment

OpenAI
Research Intern

May–Aug. 2026

Amazon
Applied Scientist Intern

May–Aug. 2025

University of Illinois Urbana-Champaign
Research Assistant

2024–Present

Republic of Korea Air Force
Cybersecurity Specialist Officer, Captain

2017–2024

Research full list

SEC-bench Pro: Can Language Models Solve Long-Horizon Software Security Tasks?

Hwiwon Lee, Jiawei Liu, Dongjun Kim, and 4 more authors

arXiv preprint arXiv:2605.26548, 2026

Adopted by OpenAI for evaluating latest models + Google VRP $20,000 bounty

Abs arXiv

SEC-bench Pro evaluates whether frontier models can hunt real vulnerabilities in large, complex software systems and reproduce working proof-of-concept inputs. The benchmark contains 344 validated vulnerabilities across the V8 and SpiderMonkey browser engines and the Linux kernel, spanning memory-safety, sandbox, JIT, race-condition, and kernel-subsystem bugs.
Agentic Vulnerability Reasoning on COTS Binaries

Hwiwon Lee, Jongseong Kim, and Lingming Zhang

arXiv preprint arXiv:2605.05000, 2026

39 zero-days (23 assigned CVEs) and $200,000+ in Microsoft bounty awards

Abs arXiv

SLYP is an end-to-end agentic pipeline for vulnerability discovery and validation on commercial off-the-shelf binaries. It combines extensible tool servers for binary exploration and dynamic debugging, and validates candidates by synthesizing debugger-verified proof-of-concept crashes. SLYP has uncovered 39 zero-day vulnerabilities in Windows systems, including 23 assigned CVEs.
SEC-bench: Automated Benchmarking of LLM Agents on Real-World Software Security Tasks

Hwiwon Lee, Ziqi Zhang, Hanxiao Lu, and 1 more author

In Proceedings of the Thirty-Ninth Annual Conference on Neural Information Processing Systems (NeurIPS’25), 2025

Abs HTML PDF

SEC-bench is a fully automated benchmarking framework for evaluating language-model agents on authentic security-engineering tasks, including proof-of-concept generation and vulnerability patching.
BENZENE: A Practical Root Cause Analysis System with an Under-Constrained State Mutation

Younggi Park, Hwiwon Lee, Jinho Jung, and 2 more authors

In 45th IEEE Symposium on Security and Privacy (Oakland’24), 2024

IEEE Symposium on Security and Privacy 2024 Distinguished Paper Award

Abs HTML PDF

BENZENE is a practical end-to-end root-cause analysis system that uses under-constrained state mutation to generate crashing and non-crashing behaviors for effective and efficient diagnosis.

News

May 26, 2026	Our new preprint, SEC-bench Pro, evaluates frontier models on long-horizon vulnerability discovery across browser engines and the Linux kernel.
May 18, 2026	Joined OpenAI as a Research Intern for summer 2026.
May 06, 2026	Our new preprint, Agentic Vulnerability Reasoning on COTS Binaries, introduces SLYP for end-to-end vulnerability discovery and validation on production binaries.
Apr 01, 2026	Recognized by Microsoft as a Most Valuable Security Researcher for 2026 Q1, marking four consecutive quarterly recognitions.
Sep 22, 2025	Awarded $10k in credits from OpenAI for my research on AI for Cybersecurity