AdvSCanner: Generating Adversarial Smart Contracts to Exploit Reentrancy Vulnerabilities Using LLM and Static Analysis

AGEStaticAGEStatic is an innovative project aimed at enhancing the security of Ethereum smart contracts by automatically generating exploit smart contracts. The project leverages large language models (LLMs) and static analysis to automatically generate adversarial smart contracts (ASCs) designed to exploit reentrancy vulnerabilities in victim contracts, which are among the most critical security issues in smart contracts.<b>Dataset</b>We have collected and integrated multiple smart contracts with reentrancy vulnerabilities from various sources. To obtain more representative samples, we filtered out ineligible and duplicate smart contracts according to the standards mentioned above, resulting in a total of 78 unique smart contracts (14 are duplicate.)<b>Size</b>: The dataset includes 78 smart contracts (14 duplicates), each verified for relevance and uniqueness,such as ERAP, ESC, Smartbugs, RSD, ATR, and SSE.<b>Standards for Dataset Collection</b>:<b>Solidity Smart Contract</b>: The AGEStatic tool we designed is aimed at Solidity smart contracts, with Solidity versions ranging from 0.4.0 to 0.8.25.<b>Open-source and Peer-reviewed Dataset</b>: The reentrancy vulnerabilities datasets are collected from widely-used or peer-reviewed open-source datasets that have obtained general public acceptance and applications in relevant research.<b>Marked as Reentrancy Vulnerability</b>: The most vital standard requires the existence of reentrancy vulnerability, which can be categorized into two types: manually injected vulnerability (MI) and real-world vulnerability (RW).<b>Detection by Static Analysis Tool</b>: These contracts in the dataset should be identified as reentrancy vulnerability by traditional static analysis tools that output reentrancy reports for each contract.<b>Fully Functional Characteristics</b>: Smart contracts with only partial functions cannot support attack verification experiments; therefore, the contracts satisfy logical integrity and full functionality characteristics.<b>Physical Experiment</b>This section describes the environment and code used for running the static analysis experiments and generating exploit contracts.<b>Static Analysis</b>: The static analysis experiments, obtained from GitHub, are run on an Ubuntu 22.04 system with the following hardware specifications:<b>Operating System</b>: Ubuntu 22.04<b>CPU</b>: Intel(R) Core(TM) i7-9750H @ 2.60GHz (2 cores and 2 threads)<b>Cache Size</b>: 12288 KB<b>Memory Size</b>: 6085248 KB<b>Exploit Contract Generation</b>: We leverage APIs of gpt-3.5-turbo, gpt-4, or gpt-4o using Python. The environment specifications are as follows:<b>Required Packages</b>:<code><strong>python==3.10.0</strong></code><code><strong>openai==0.28.0</strong></code><code><strong>py-solc-x==2.0.2</strong></code><b>Experiment Results</b>The experimental results include RQ1, RQ2, RQ3, and RQ4.