Interactive smart contract auditing with ChatGPT

Abstract

In this paper, we will analyse the efficacy of a web-based smart contract auditor that uses LLMs to audit smart contracts in Solidity. First, we propose a set of pre-defined prompts to help smart contract developers query the LLM. Next, we allow users different follow-up queries to harness the unique language capabilities of GPT, improving the detection and user experience of the auditing tool. Our solution: AuditGPT is a web-based interface that allows users to select different prompting styles for their code. Our experiments show that flexibility in prompting styles can aid human developers in understanding code vulnerabilities as well as improved detection rates compared to trivial vulnerability checking