Reducing Compliance Violations in Ethereum Smart Contracts: A Multi-Agent LLM Approach to ERC Standard Auditing

Abstract

Ethereum is a decentralized blockchain platform that allows developers to deploy and run smart contracts, which are self-executing programs responsible for handling digital transactions without intermediaries. ERC standards define how these smart contracts are expected to behave in the Ethereum ecosystem. When these rules are not implemented correctly, contracts may contain security weaknesses that can lead to financial loss or unexpected behavior. For this reason, verifying whether a contract complies with ERC requirements is an important task during the development process. However, compliance verification is still often performed manually, which makes the process slow and dependent on expert knowledge. Most existing static analysis tools mainly detect predefined vulnerability patterns, but they may miss behavioral deviations from Ethereum Request for Comments (ERC) specifications that are not explicitly encoded as patterns. In this study, we present a multi-agent LLM framework designed to automate ERC compliance auditing. The system extracts contract-specific code fragments and evaluates them using multiple independent Large Language Model agents. Their outputs are aggregated through a confidence-weighted mechanism that aims to stabilize the final decision. Experiments on ERC-20, ERC721, and ERC-1155 contracts show that the multi-agent configuration improves recall and reduces false negatives compared to single-agent auditing.