无源码场景下智能合约字节码漏洞检测技术综述
首发时间:2026-03-05
摘要:智能合约漏洞频发且损失巨大。然而,现实中绝大多数合约仅以无源码的EVM字节码形式存在,致使传统依赖源码的分析工具失效。针对此“无源码”场景,本文全景回顾了2021-2026年间字节码漏洞检测的前沿研究。文章首先剖析了字节码的底层特征与语义鸿沟;其次,从静态分析、动态模糊测试、深度学习/图神经网络、大模型反编译四大流派,系统梳理了技术演进脉络;接着,指出现有技术在反汇编精度与业务逻辑理解等方面的核心痛点;最后,展望了基于大语言模型(LLM)与检索增强生成(RAG)重构高维语义图的未来范式。
关键词: 智能合约安全 字节码分析 漏洞检测 反编译 大语言模型
For information in English, please click here
Survey on Smart Contract Bytecode Vulnerability Detection Techniques in No-Source Scenarios
Abstract:Smart contract vulnerabilities often lead to massive economic losses. However, most real-world contracts exist only as closed-source EVM bytecode, rendering traditional source-code-dependent analysis tools ineffective. Focusing on this "no-source" scenario, this paper comprehensively reviews top-tier research on bytecode vulnerability detection from 2021 to 2026. First, it analyzes bytecode's low-level characteristics and the semantic gap. Second, it systematically categorizes the evolution of detection techniques into four major schools: static analysis and symbolic execution, dynamic fuzzing, deep learning and graph neural networks, and large language model (LLM)-driven decompilation. Third, it dissects the common pain points of existing methods regarding disassembly precision and business logic understanding. Finally, it envisions the future trend of reconstructing high-dimensional semantic graphs based on LLMs and Retrieval-Augmented Generation (RAG).
Keywords: smart contract security bytecode analysis vulnerability detection decompilation large language model
基金:
引用

No.****
同行评议
勘误表
无源码场景下智能合约字节码漏洞检测技术综述
评论
全部评论