Make Agent Defeat Agent: Automatic Detection of Taint-Style Vulnerabilities in LLM-based Agents

Fengyu Liu, Yuan Zhang, Jiaqi Luo, Jiarun Dai, Tian Chen, Letian Yuan, Zhengmin Yu, Youkun Shi, Ke Li, Chengyuan Zhou, Hao Chen, Min Yang

34th USENIX Security Symposium · Day 1

The rapid proliferation of Large Language Models (LLMs) has led to a new paradigm in software development: **LLM-based agents**. These AI-powered applications are designed to understand natural language instructions, perceive external environments, and intelligently execute complex tasks, often involving critical operations like code execution and sensitive data handling. However, this transformative technology introduces novel security challenges. Recent research highlights that these agents are highly susceptible to **taint-style vulnerabilities**, a class of flaws where malicious user input flows into security-sensitive operations without proper sanitization, potentially leading to severe consequences such as remote agent takeover, information leakage, and arbitrary code execution.

AI review

Solid, practical security research that actually ships CVEs. AgentFuzz addresses a real gap — static analysis chokes on Python's dynamic chaos and LLM-mediated control flow — and the 34 zero-days across popular agents like AutoGPT prove the approach works. Not paradigm-shifting, but genuinely useful work that will make agent developers uncomfortable in exactly the right ways.