SVACS : Smart Contract Vulnerability Analysis via Control Flow Semantics
IEEE Access · 2026 · 14 · 123386-123397
DOI: 10.1109/access.2026.3722527 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Ankur Jain und 2 weiteren Autor:innen. Die Arbeit nennt 28 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Smart contracts play a key role in blockchain application development, but vulnerable smart contracts cause serious security risks to blockchain ecosystems. While most of the existing vulnerability detection techniques rely on analyzing Solidity source code, over 94% of deployed smart contracts are not open source. These smart contracts can only be accessible in bytecode form so these tools are not applicable. Existing bytecode based tools also have some limitations include improper handling of long opcode sequences and inability to detect multiple co-existing Smart Contract Weakness Classification (SWC) vulnerabilities. In this paper, we propose SVACS, a bytecode-based analysis framework to identify multiple co-existing vulnerabilities based on SWC registry to align with industry-standard security guidelines and assist security reviewers to ensure smart contract security. SVACS represents an Ethereum smart contract as control-flow graph (CFG), where each basic block is encoded using GraphCodeBERT. Block embeddings are refined using residual graph attention networks (RGAT) layer to capture inter-block control-flow dependencies. Contract-level vulnerability predictions are obtained by aggregating block-level logits. This enables SVACS to detect multiple co-existing SWC identifiers simultaneously. Performance is evaluated on a real-world dataset of 5,478 Ethereum contracts comprising 10 SWC vulnerability identifiers. The labels are obtained by combining a manually audited Consolidated Ground Truth corpus and additional smart contracts collected from Etherscan API and labelled using Slither-based analysis. To handle the multi-label and imbalanced nature of the dataset, iterative stratified splitting is used to maintain label distributions across splits, and label-wise thresholding is applied during prediction. It is found that SVACS outperforms state-of-the-art tools with 87.33% accuracy, 94.69% TPR, 15.67% FPR, 5.31% FNR, and 81.26% F1-score. SVACS achieved good performance across all 10 SWC vulnerability identifiers, with TPR ranging from 75.00% to 99.50%, FPR ranging from 1.87% to 88.81%, FNR ranging from 0.50% to 25.00%, and F1-score ranging from 54.04% to 92.89%. SVACS uses the GraphCodeBERT encoder and performs better than both CodeT5 and CodeT5p-220M in terms of higher accuracy by 8.19 and 7.89 percentage points, higher TPR by 1.37 and 2.86 percentage points, lower FPR by 10.54 and 9.76 percentage points, lower FNR by 1.37 and 2.86 percentage points, and improved F1-score by 10.21 and 9.65 percentage points, respectively. By performing directly on Ethereum Virtual Machine (EVM) bytecode and designing control-flow structure, SVACS offers a more effective solution for real-world smart contract security analysis.
Zitieren
APA
Jain, A., Anand, A., & Tripathy, S. (2026). SVACS : Smart Contract Vulnerability Analysis via Control Flow Semantics. IEEE Access, 14, 123386-123397. https://doi.org/10.1109/access.2026.3722527
Vancouver
Jain A, Anand A, Tripathy S. SVACS : Smart Contract Vulnerability Analysis via Control Flow Semantics. IEEE Access. 2026;14:123386-123397. doi:10.1109/access.2026.3722527
BibTeX
@article{Jain2026,
author = {Jain, Ankur and Anand, Abhishar and Tripathy, Somanath},
title = {SVACS : Smart Contract Vulnerability Analysis via Control Flow Semantics},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {123386--123397},
doi = {10.1109/access.2026.3722527}
}
EUVIMED