Improving Smart Contract Analysis with Large Language Models: The SLLM System

Abstract

Smart contracts are self-executing agreements stored on the blockchain, widely used in finance, supply chain, IoT, and other fields. However, vulnerabilities in smart contract code can be exploited, leading to irreparable losses. Numerous smart contract vulnerability detection tools have been developed, with Slither being a prominent open-source static analysis tool. Despite its strengths, Slither suffers from false positives and false negatives. This paper introduces the SLLM system (Synergy of Slither and Large Language Models), which combines Slither with large language models (LLMs) like GPT-4. By leveraging pseudocode conversion and result feedback enhancement mechanisms, the system improves the accuracy of vulnerability detection. Experimental results show that SLLM outperforms Slither in reducing both false positives and false negatives across various common vulnerability types, demonstrating the effectiveness of integrating LLMs in smart contract static analysis.