Smart Contract Vulnerability Detection Based on Prompt-guided ChatGPT

Abstract

Millions of smart contracts are deployed on various blockchain platforms, involving extensive digital assets. However, vulnerabilities within these smart contracts have resulted in substantial exploitation and asset losses. Traditional methods for detecting smart contract vulnerabilities are limited by their narrow detection range and enormous computational cost. This paper investigates how large language models (LLMs), particularly ChatGPT 4, can be leveraged to detect vulnerabilities in smart contracts. We conduct a comprehensive survey of several existing detection methods for smart contract vulnerabilities. Meanwhile, we design a variety of prompt information, and added contract opcodes and expert rules as auxiliary information. Utilizing ChatGPT, we evaluate the effectiveness of the large language model in identifying vulnerabilities across two datasets. The experimental results demonstrate that ChatGPT, informed by specific prompts, can effectively pinpoint vulnerabilities, highlighting the utility of LLMs in enhancing the security of smart contracts.