LLM-Based Tooling for Smart Contract Auditing

Abstract

Traditionally, smart contract auditing has been conducted using analysis tools and manual review processes. However, these tools often struggle to detect complex vulnerabilities and novel attack vectors. Advancements in Large Language Models (LLMs) have introduced new possibilities for enhancing smart contract audits by using their contextual understanding and reasoning capabilities. This paper provides a review of existing LLM-based smart contract auditing tools. We analyze key methodologies, strengths, and limitations of six such tools. While LLM-based tools demonstrate significant potential in detecting complex vulnerabilities, challenges such as false positives and token length limitations persist. Our comparative evaluation highlights performance differences, showcasing the potential of LLMs to complement traditional auditing tools. Finally, we discuss current challenges and future directions for improving LLM-based auditing, aiming to enhance security in blockchain ecosystems.