Miriam W. Kaara, Jael S. Wekesa, Michael W. Kimwele
The integrity and credibility of educational institutions worldwide are being undermined by the growing issue of academic certificate fraud. The public's trust in the educational system, the integrity of job opportunities, and the legitimacy of authentic credentials are all negatively impacted by academic dishonesty. The majority of academic credentialing systems, both centralized and blockchain-based, focus on academic credential validation and immutability, even though academic institutions have embraced technology to improve academic credentialing procedures. This suggests that their support for intelligent and private fraud detection is minimal. In this work, we suggest an integrated method that combines smart anomaly detection with Federated Learning (FL). This makes it possible for several organizations to build a model for detecting certificate fraud without disclosing private information to other organizations. Each institution's private information is retained and incorporated into a global model. This enables the framework to handle concerns about data ownership and privacy as well as regulatory compliance. The framework uses Extreme Gradient Boosting (XGBoost) to identify anomalies in metadata and Convolutional Neural Networks (CNNs) to detect visual forgeries in certificates. A decentralized node is used to train the models, which are merged using the Federated Averaging (FedAvg) algorithm. The federated model performs well in anomaly detection, according to the experiments conducted. When compared to conventional centralized approaches, it significantly reduces false positives and false negatives, with an accuracy of up to 94% and an AUC of 0.97. The model's findings imply that a federated learning approach would make it possible for institutions to detect fraud in a secure, scalable, and cooperative manner. In this regard, this framework offers a workable way to create digital credential systems that are more trustworthy and protect privacy. Key words: Federated Learning, Blockchain, certificate fraud detection, privacy preservation, distributed datasets, secure collaboration.
The rapid digitization of financial services has created increasingly complex environments in which financial institutions must process large volumes of heterogeneous transaction data while simultaneously protecting customers, detecting fraud, managing financial risks, and complying with regulatory requirements. Traditional centralized and rule-based fraud detection systems face significant challenges associated with data volume, processing latency, evolving fraudulent behaviors, class imbalance, and the increasing sophistication of cyber-enabled financial crimes. This paper proposes a distributed intelligent analytics framework for blockchain-based fraud detection and risk management in financial institutions. The framework integrates distributed big data analytics, artificial intelligence, machine learning, blockchain, graph-based learning, and intelligent decision support into a unified architecture. Distributed computing provides scalable processing of heterogeneous financial datasets, while artificial intelligence identifies anomalous transactions and predicts potential risks. Blockchain provides a complementary integrity, traceability, and verification layer for financial transactions. Graph Neural Networks can further model relationships among customers, accounts, devices, merchants, and transactions, enabling the detection of complex fraud patterns that may not be visible through transaction-level analysis. The framework builds on Ramareddy's work on distributed big data analytics for scalable knowledge discovery in heterogeneous systems and Chhunchha's investigation of blockchain's influence on financial institutions. Recent research also indicates growing interest in machine learning, graph-based models, federated learning, and blockchain for financial fraud detection. The proposed framework addresses important challenges including scalability, privacy, class imbalance, concept drift, explainability, cybersecurity, and regulatory compliance. The paper argues that combining distributed analytics with blockchain and AI can provide financial institutions with a more scalable, transparent, adaptive, and intelligent approach to fraud prevention and financial risk management.
Pipelines that pair a large language model with a static analyzer, feeding findings back as repair instructions, appear throughout recent smart contract repair research. They rest on a rarely examined assumption: that the analyzer output serving as the oracle faithfully records what the analyzer found. I report three ways that assumption fails, identified during a four-contract instrument-validation exercise preceding a planned repair study. First, Mythril v0.24.8 can exit without reaching the analysis phase while returning exit status zero, empty standard error, and a findings array byte-identical to that of a genuinely clean scan; the failure is reported in a sibling JSON field that finding-extraction code has no reason to read. Second, 12 of 23 Slither findings in my validation set fell outside the high, medium, and low impact bands, so an unfiltered count measures a composite whose components may not behave alike under repair. Third, keying finding identity on source location breaks across repair rounds. On the one contract carried through three rounds, location-based keying inflated resolved findings from 7 to 12 and introduced findings from 2 to 7. The underlying instability is established in the warning-tracking literature; my contribution is its consequence for repair metrics, where it biases both transition counts upward and can confound comparison between methods producing differently sized .patches. I separately report an executed exploit showing a specification-level authorization defect that produced no high or medium impact finding. I propose calibration procedures for each hazard and release the harness, contracts, and raw analyzer output at doi:10.5281/zenodo.21586404.
This paper presents the Auditable Zero-knowledge Transformer (AZT) framework for privacy-preserving and auditable tax fraud detection. AZT combines transformer-based anomaly detection with zero-knowledge proof (ZKP) verification so that a tax authority or regulator can verify fraud-detection outcomes without accessing sensitive taxpayer records or proprietary model parameters. The framework is scalable in the specific sense of low-latency audit verification: proof verification remains sub-second, whereas proof generation is intentionally performed asynchronously after local inference. Model integrity is enforced through Merkle-root commitments to authority-approved parameters, and the ZKP statement proves that the committed transformer was executed correctly and that the resulting risk score satisfies a public audit threshold. Experiments on UCI-TFD, IRS-Pub, and CorpPay compare AZT with classical machine-learning baselines, including Random Forest and XGBoost, and with an equivalent plaintext transformer. Detection quality improves by up to 5.3% in F1-score over classical machine-learning baselines, while the circuit-compatible AZT inference incurs only about 0.5% F1-score degradation relative to the plaintext transformer baseline. Overall, this work advances secure AI for digital governance by integrating modern deep learning with cryptographic verification, offering a practical foundation for fraud-detection systems in which transparency and confidentiality must be satisfied simultaneously.
The rapid digitalization of financial services has transformed the global financial ecosystem, enabling faster transactions, enhanced customer experiences, and greater financial inclusion. However, this digital transformation has simultaneously increased the complexity, scale, and sophistication of financial fraud. Traditional rule-based fraud detection systems often struggle to identify evolving fraud patterns, resulting in delayed responses, increased false positives, and substantial financial losses. Artificial Intelligence (AI)-powered real-time fraud monitoring systems have emerged as a transformative solution capable of detecting suspicious activities instantly through advanced data analytics, machine learning, deep learning, natural language processing, and behavioral intelligence. These systems continuously analyze vast volumes of transactional and non-transactional data, enabling financial institutions to identify anomalies, predict fraudulent behavior, and automate risk management processes with unprecedented accuracy and speed. This literature review examines the evolution, applications, technological foundations, benefits, challenges, and future directions of AI-powered real-time fraud monitoring systems in modern financial services. The review highlights how AI enhances fraud detection capabilities across banking, payment systems, insurance, digital wallets, cryptocurrencies, and investment platforms while discussing critical concerns related to privacy, algorithmic bias, explainability, cybersecurity, and regulatory compliance. The findings demonstrate that AI-driven fraud monitoring represents a fundamental component of modern financial security infrastructure and will continue to shape the future of fraud prevention in increasingly digital financial environments.
The increasing adoption of cryptocurrencies has created new opportunities for digital financial innovation while simultaneously exposing individuals and institutions to sophisticated forms of financial fraud. Conventional rule-based fraud detection systems have become inadequate in addressing the dynamic and complex nature of blockchain-enabled financial crimes, leading to growing interest in the application of artificial intelligence (AI). This study systematically reviews the literature on artificial intelligence techniques for cryptocurrency fraud detection, with particular emphasis on their relevance to the Kenyan digital financial ecosystem. The review was conducted using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 framework. Peer-reviewed studies published between 2020 and 2026 were identified from Scopus, Web of Science, IEEE Xplore, ScienceDirect, SpringerLink, and Google Scholar. Following the screening and eligibility assessment, 19 studies were included in the final qualitative synthesis. The findings reveal that machine learning, deep learning, hybrid AI models, and blockchain analytics significantly enhance cryptocurrency fraud detection by improving anomaly detection, transaction monitoring, predictive accuracy, and anti-money laundering compliance. Compared with traditional rule-based approaches, AI techniques provide faster, more adaptive, and scalable solutions capable of detecting evolving fraud patterns in decentralized financial systems. However, the review also identifies challenges relating to limited high-quality datasets, algorithmic bias, lack of explainability, cybersecurity risks, privacy concerns, and inadequate regulatory frameworks, particularly within developing economies. Furthermore, the review highlights a scarcity of empirical research focusing on cryptocurrency fraud detection in Kenya and identifies opportunities for developing localized datasets, explainable AI models, and context-specific regulatory frameworks. The study concludes that artificial intelligence has considerable potential to strengthen cryptocurrency fraud detection and financial security in Kenya, provided that technological, ethical, and regulatory challenges are adequately addressed. The findings provide valuable insights for researchers, financial institutions, technology developers, and policymakers seeking to enhance AI-driven fraud prevention within the country's evolving digital financial ecosystem.
Alternative finance platforms, including crowdfunding, peer-to-peer lending, equity-based platforms, and token-based fundraising mechanisms, have become important channels for financing entrepreneurial, social, and investment-oriented initiatives. Yet their reliance on digital intermediation, dispersed participation, and information asymmetry creates opportunities for fraud, undermining trust, investor protection, and platform sustainability. This study provides a systematic review of fraud detection and prevention in alternative finance, with crowdfunding emerging as the most extensively represented empirical domain. Methodologically, the paper combines a PRISMA-guided systematic literature review with a hybrid topic-modeling strategy that integrates neural topic modeling and probabilistic refinement, thereby supporting both transparent corpus selection and data-driven thematic synthesis. The findings show that Artificial Intelligence (AI), Machine Learning (ML), Natural Language Processing (NLP), and blockchain-based mechanisms are recurrently discussed as promising tools for detecting, preventing, or mitigating fraud. AI and ML approaches are mainly used to identify anomalies, suspicious textual patterns, behavioral signals, and transaction irregularities, while blockchain-based approaches are associated with transparency, traceability, smart contracts, and conditional fund release. The review also shows that fraud differs across alternative finance models, ranging from campaign misrepresentation and intentional and premeditated non-delivery in crowdfunding to borrower or platform misreporting in lending-based models and misleading disclosures or white-paper manipulation in ICO/STO contexts. A central challenge across the literature is the scarcity of labeled fraud data, which limits the use and benchmarking of supervised ML models. Overall, this study contributes by linking a reproducible hybrid SLR methodology to a structured synthesis of fraud types, platform-specific vulnerabilities, and AI-, ML-, and blockchain-based detection strategies in alternative finance.
As the growth of the FinTech platforms continues, there is an increasing demand for intelligent, secure and traceable solutions that can provide real-time detection of fraudulent transactions and shield financial records from manipulation. In this research, an Artificial Intelligence-powered blockchain framework, combining machine learning for fraud detection and permissioned blockchain for validation, was proposed. It was found that ensemble models performed better than a linear baseline. The overall best balance of precision, specificity and F1 score was obtained with the Random Forest model, and the highest precision–recall was obtained with the Extra Trees model, with fraud recall slightly better. In addition, feature-importance analysis revealed a small number of transaction attributes, which were anonymised, that most significantly affected fraud classification. The chosen model was then connected to a prototype of a chained hash blockchain that preserved the hashes of transactions, the time, the predicted probability of fraud, the validation result, and the version of the model. Through hash inconsistency, the prototype was able to detect any transaction modifications which might have been made on purpose and successfully ensured ledger integrity. The results illustrate how both AI and blockchain technologies complement each other. AI is effective in detecting fraud accurately and on time, and blockchain enhances the traceability, auditability and tamper resistance of transactions.
Smart contract vulnerability detection requires evaluation protocols that separate real representation signal from dataset-specific artifacts. DIVE provides lifecycle-based tabular features for Ethereum smart contracts, but benchmark performance alone cannot show whether a dominant feature group is useful or only benefits from having many columns. This study examines Opcode Distribution features using 22,330 contracts, 397 processed features, and eight DASP-aligned vulnerability labels. Five multi-label learning configurations were evaluated under 3 x 5 repeated cross-validation, followed by global feature-group ablation, size-controlled random opcode ablation, per-label degradation analysis, cumulative stability analysis, and opcode-profile group-aware robustness checking. MultiOutput LightGBM achieved the best baseline performance, with Micro-F1 of 0.91396, Macro-F1 of 0.82464, and Macro-PR-AUC of 0.90146. Removing the full Opcode Distribution group reduced Macro-F1 to 0.78745, while removing a same-sized random opcode subset produced Macro-F1 of 0.82404. The findings indicate that Opcode Distribution acts as a collective predictive representation rather than a feature-count artifact, without implying causal vulnerability mechanisms.
One of these financial crimes, which seem to sound like a concept straight out of a dream until you get a sense of the magnitude of the issue, is money laundering. According to the United Nations, Between $800 billion and $2 trillion in illicit money is transacted through the world financial system each and every year. The problem with this approach is that the criminals seldom use only one bank. They thread their way across five, ten, and sometimes dozens of institutions, all seeing merely a harmless nugget. In isolation, looking at his or her own transaction logs, no single bank will easily know that there is a problem. This paper is about a system, called AMLNet, which tackles this blind spot. Unlike the traditional approach, which would allow banks to share their customers' data with each other,AMLNet trains a detection model on customers' data within each bank, and shares only what the detection model learned from the data, not the data itself. All collaborative training is documented in a blockchain ledger, making it transparent and tamper-proof. With a Zero-Knowledge Proof, each bank is able to prove cryptographically that it is acting honestly, but not disclose anything private. A graph of transaction data (accounts as nodes, transfers as edges) is used to extract structural features, which are compressed by PCA before being input to a Multi-Layer Perceptron (MLP) risk-scoring classifier of each account. Together they increase fraud recall by approximately 20% over any single institution operating alone, while maintaining a low false positive rate, and that the overall computation time is less than 10 minutes on an average laptop.
As the blockchain technology and decentralized finance have grown rapidly, the number of fraudulent and anomalous activities has risen.The paper suggests a detectable graphbased anomaly detection system to detect suspicious Ethereum transactions.One 10,000 Ethereum transactions dataset was gathered through the Etherscan API within a 14 hour observation period and a directed transaction graph was created out of that dataset, where 14 behavioral node features were engineered.Three graph neural network (GNN) models, namely, Graph Convolutional Network (GCN), Graph Attention Network (GAT), and GraphSAGE, were checked on 5-fold cross-validation, and compared to three standard baseline classifiers, which are Logistic Regression, Random Forest, and XGBoost.GraphSAGE had the highest overall accuracy of 82.32, F1-score of 0.6389, and ROC-AUC of 0.8202, and GCN and GAT had near-zero recall on the minority class.XGBoost was the best baseline with the highest accuracy (94.41) but with significantly lower recall (0.2766) and F1-score (0.3801) compared to GraphSAGE, which is indicative of graph-based models being more balanced in precision and recall in detecting anomalies with class imbalance.The Local Interpretable Model-agnostic Explanations (LIME) showed outgoing transaction value features and account balance to be most important predictors of anomalous behavior.The results establish the promise of using GNNs in conjunction with explainable AI to secure blockchains, as well as reveal the challenges such as the class imbalance and ground-truth verified labels.
The complete codebase and supplementary materials for this study have been archived on Figshare to ensure full reproducibility and to facilitate adoption by other researchers and practitioners. The archive includes all Python scripts used for data preprocessing, model training, hyperparameter tuning, threshold optimisation, and SHAP explainability analysis. Also included are the processed CSV files used for the analysis, along with all figures and tables presented in this paper. The repository is organised to enable straightforward replication of the experiments and adaptation of the framework to other datasets or blockchain platforms.
Hai Liang, Xiaoye Lu, Changsong Yang, Yujue Wang · 6 authors
Smart contracts are immutable programs that automatically execute predefined logic. Once deployed, their underlying vulnerabilities are notoriously difficult to patch and highly susceptible to malicious exploitation, often leading to severe financial losses. Although existing vulnerability detection methods have demonstrated certain advantages, they still fail to achieve adequate structural–semantic coverage of vulnerability-relevant behaviors, as they are unable to jointly model opcode semantics, control-flow transitions, and data-dependency relations. To overcome these limitations, this paper proposes a novel smart contract vulnerability detection model named Cross-aligned Penetrative Graph Network (CPGNet). Specifically, CPGNet first constructs control flow graphs and data flow graphs from the abstract syntax tree, and combines them with opcode semantic embeddings to form a multidimensional initial code representation. Based on this representation, a cross-alignment mechanism is introduced to effectively capture and integrate the complex interactions between control-flow transitions and data-flow dependencies. Furthermore, an explicit–implicit feature penetration architecture is designed to inject shallow local opcode patterns into the deep semantic modeling process, enabling multi-source features to dynamically complement each other. By jointly modeling opcode semantics, control-flow structures, and data-dependency relations, CPGNet significantly enhances the representation capability for hidden and complex vulnerability patterns. Experimental results on two datasets show that CPGNet achieves stable performance, with F1-scores of 88.69% and 90.58% on the benchmark Ethereum dataset, and 78.10% and 71.53% on DIVE for reentrancy and timestamp dependency detection, respectively. These results verify the effectiveness of jointly modeling opcode semantics and graph-level structural dependencies.
Since the first implementation of a blockchain with Bitcoin in 2009, cryptoassets created and transacted using blockchain technologies have grown and diversified significantly. Because regulatory regimes, which govern cryptoassets, do not have global coverage, criminal actors find opportunities to commit cryptoasset fraud. While it can be difficult to distinguish between cryptoassets that are honest but high risk and cryptoassets that are outright fraudulent, investors seeking significant returns frequently invest in unregulated cryptoassets, namely cryptocurrencies and non-fungible tokens (NFTs). This study provides a crime script analysis to examine the chronological and functional steps offenders use to execute cryptoasset fraud. It considers three types of crypto asset fraud and how they have functioned over time: Ponzi schemes, cryptoasset exit scams, such as cryptocurrency “rug pulls,” and NFT “mint-and-run” schemes, where invested value is stolen from a crypto asset project. By outlining the fundamental crime script of cryptoasset fraud, this study considers the implications for regulators. Of note, this study shows that while the stages of cryptoasset frauds are consistent, the speed at which frauds are executed has, on average, increased significantly. This rapidity of execution provides enduring challenges to regulators, who often cannot respond quickly. This challenge must be considered if regulation is to be effective.
The growth of the usage of decentralized applications on Ethereum has seen the rise of an increasing number of bad actors that are using it to commit fraud, phishing, money laundering and financial scams. Traditional detection methods are less effective to detect accounts with more complex and changing behaviours. The paper suggests a novel multi-level framework for detecting malicious Ethereum accounts based on supervised classification and adaptive anomalous account verification using a routing based on probabilities. To get the transaction behavior features, opcode features and time-interval features from the publicly available EtherShield data set, we use the entire data set to extract the entire features. The first is a Level-1 where an XGBoost machine learning model classifies Ethereum accounts into Fair, Likely Malicious and Malicious categories, and outputs calibrated probability scores for any account. Uncertainty about malicious accounts are escalated to level-2 where anomaly verification and behavioural re-assessment are carried out by using models such as Random Forest and Isolation Forest. The final classification is obtained by decision fusion process, which combines the results obtained from both levels. The Random Forest-based verification module is evaluated in the experiments and is found to be 95% accurate, 94% macro precision, 95% macro recall and 95% macro F1-score, which is significantly better than the Isolation Forest (baseline). Besides, stratified 5-Fold Cross Validation further demonstrates that the proposed framework is robust and generalizable with a mean accuracy of 94.96% ± 0.43, mean precision of 94.40% ± 0.42, mean recall of 94.84% ± 0.47 and mean F1-Score of 94.66% ± 0.42. The proposed framework proves to be an effective solution to minimize misclassification, enhance the reliability of detection and offer a scalable answer to safeguard Ethereum blockchain ecosystems from newly emerged malicious activities.
Decentralized Finance (DeFi) has revolutionized financial services by eliminating traditional intermediaries, but this openness creates new vulnerabilities that malicious actors exploit for fraud. The pseudonymous nature of blockchain transactions and lack of centralized oversight make traditional fraud detection methods inadequate for the DeFi ecosystem. This paper introduces ChainGuard, an end-to-end fraud detection system that leverages verifiable machine learning with zero-knowledge Succinct Non-Interactive Arguments of Knowledge (zk-SNARKs). ChainGuard utilizes a comprehensive approach that combines advanced feature extraction from Ethereum blockchain transaction data, optimized machine learning models, and on-chain verification through zk-SNARKs. Our solution enables privacy-preserving fraud detection while maintaining the ability to verify results without exposing sensitive transaction data and the internal architecture of the model. We demonstrate that ChainGuard achieves permissible accuracy in detecting fraudulent activities across Ethereum and various DeFi platforms while ensuring computational efficiency through multiple optimization techniques, including quantization. Experimental results show that our approach achieves performance comparable to traditional fraud detection methods while maintaining the decentralized and trustless nature of blockchain systems.
Sabri Hisham, Mokhairi Makhtar, Azwa Abdul Aziz, Ahmad Firdaus · 6 authors
Abstract The global expansion of blockchain technology has unfortunately been accompanied by a rise in fraudulent activities within decentralized applications (DApps), leading to substantial financial losses. The immense volume of transaction data (big data) makes manual detection of abnormal account behavior impossible, necessitating the use of automated machine learning (ML) techniques. Existing anomaly machine learning detection approaches often rely on single-classifier models that suffer from limited generalization, high false-positive rates, or insufficient feature relevance, thereby compromising detection accuracy and system security. Moreover, the high dimensionality and complexity of blockchain data necessitate more sophisticated and robust methodologies that can effectively identify relevant features and leverage the strengths of multiple learning algorithms. This study addresses a key gap by proposing a novel anomaly detection framework for the Ethereum blockchain that distinctively integrates the Boruta feature selection algorithm with a combination of ensemble methods and a fuzzy logic classifier. Specifically, we investigate the performance of various ensemble techniques (bagging, boosting, voting, and stacking) combined with foundational models (Decision Tree, Random Forest, K-Nearest Neighbors, and XGBoost), including a specialized Fuzzy ENORA model. The objective is to significantly enhance the accuracy of anomaly detection. Our results demonstrate that the ensemble models consistently and significantly outperformed single-classifier models, achieving a mean performance metric of 0.99 across accuracy, precision, recall, and F1 score, affirming the robustness of the proposed Boruta-driven ensemble approach for securing blockchain transactions.
Ethereum account classification is essential for identifying individuals engaged in illicit transactions and analyzing behavioral patterns across various account types. This process serves as a critical mechanism for monitoring and regulating unlawful activities within transactional markets. However, the Ethereum network exhibits the characteristics of a complex heterophilic graph which poses significant challenges to the effectiveness and performance of conventional graph neural networks (GNNs). To address this challenge, the present study proposes FSGCN(Fourier-Sage GCN), a novel architecture for heterophilic graph neural networks (GNNs) that integrates Kolmogorov–Arnold Networks (KANs) with GraphSAGE. FSGCN is specifically designed to adapt efficiently to the structural complexity of heterophilic graphs. By leveraging KANs to extract high-order neighborhood information and employing GraphSAGE to capture low-order neighborhood patterns, FSGCN effectively aggregates both homophilic and heterophilic features, thereby improving classification performance. Furthermore, to improve training efficiency and generalization, we propose the MLPInit weight initialization scheme and the DropEdge graph augmentation technique. Experiments on a large-scale Ethereum transaction dataset show that FSGCN achieves an F1-score of 91.8% and a classification accuracy of 91.6%, significantly outperforming traditional homophilic and heterophilic GNN baselines. Additionally, FSGCN demonstrates high training efficiency, completing each epoch in just 2.302 s per epoch and improving overall training speed by 130.4% compared to conventional GraphSAGE.
This is an independent research project with publicly released, reproducible code (not a peer-reviewed publication). We ask which class of behavioural signal drives machine-learning detection of fraudulent Ethereum accounts: graph, transaction (value/volume), or temporal (timing) features, on 9,307 labelled accounts. Crucially we distinguish degree-count graph features from true graph-topology features (PageRank, k-core, clustering, degree centrality) reconstructed from a 242,518-node, 1.65M-edge transaction graph. Transaction-value features are the strongest single class (PR-AUC 0.93), but true graph-topology significantly outperforms degree counts (PR-AUC 0.84 vs 0.70, p<1e-6) and adds the most on top of transaction features; PageRank is the single most informative feature. The topology result survives a time-respecting leakage audit (features rebuilt from each account's earliest 70% of transactions). All code, data pointers, figures, and tests are released.
High-quality smart contract auditing datasets are crucial for evaluating security tools and advancing smart contract security research. Two major limitations of existing datasets are the manual-induced scalability bottleneck and the deficiency in data granularity and diversity. To address these limitations, we propose GiANT, an automated framework designed to curate smart contract auditing datasets by distilling vulnerability insights from real-world auditing reports. GiANT employs a divide-and-conquer strategy coupled with the Chain-of-Thought technique to extract structured vulnerability information from Code4rena reports, followed by an LLM-as-a-judge mechanism to perform rigorous quality assurance. To evaluate GiANT's effectiveness, we run it on 388 real-world audit reports and generate the GiAnt Corpus comprising 7,711 vulnerability findings across five severity levels. Manual assessment of the dataset demonstrates exceptional reliability in information extraction, achieving a mean quality score of $4.76\pm0.37$ (out of 5) with inter-rater agreement $κ$ of 0.88. We further validate the practicality of our dataset by benchmarking 4 state-of-the-art LLMs on vulnerability detection, code summarization, mitigation recommendation, and automated gas optimization tasks, to establish performance baselines, thereby providing a valuable data foundation for future research in automated smart contract auditing.
Smart contract vulnerabilities in Decentralized Finance (DeFi) protocols resulted in over 1.49 billion USD in confirmed losses in 2024 alone, across 192 incidents [1]. As LLM-based vulnerability detection emerges as a promising approach to address these threats, the quality of evaluation datasets has become a critical bottleneck. Existing datasets suffer from three fundamental problems: they are built on outdated Solidity versions (e.g., v0.4) that no longer reflect modern DeFi contracts [5][6][7]; they rely on automated or LLM-generated annotations that introduce hallucination-driven label noise [9][10]; and they apply coarse single-layer labeling that fails to capture the semantic complexity of real-world business logic vulnerabilities [6][7][11][12]. We present Bastet, an expert-labeled DeFi smart contract vulnerability dataset that addresses all three problems through real-world audit findings (2021-2024), human expert annotation with discussion-based consensus, and a two-layer taxonomy of 46 Tags and 77 Subtags. Bastet comprises 4,402 findings collected from 394 Code4rena competitive audit reports spanning April 2021 to November 2024, of which 849 findings are fully annotated by white-hat security researchers from the DeFiHackLabs community. All annotations are produced through a two-annotator consensus workflow, ensuring label accuracy grounded in real-world vulnerability root causes.
Ethereum is a widely adopted blockchain platform that supports a large number of decentralized applications. Despite its rapid growth, Ethereum remains vulnerable to security threats, particularly phishing attacks that exploit transactional behavior. This study investigates the effectiveness of tree-based ensemble learning models for detecting phishing transactions on the Ethereum network using an imbalanced transaction dataset. Seven tree-based ensemble classifiers are empirically evaluated under a cost-sensitive learning framework, with performance assessed using the Matthews Correlation Coefficient (MCC) as the primary metric. The results show that boosting-based ensembles substantially outperform bagging-based approaches and a single decision tree. In particular, Gradient Boosting achieves the strongest detection performance with an MCC of 0.9742, while CatBoost provides a trade-off between detection performance and computational efficiency, achieving competitive detection accuracy with the lowest average inference time (approximately 1.54 µs per transaction). The findings demonstrate that accurate and robust phishing detection can be achieved using a compact feature representation, enabling practical deployment with reduced computational overhead.
Cryptocurrency fraud on blockchain platforms continues to cause substantial financial losses, creating an urgent need for detection systems that are not only accurate but also interpretable for operational and regulatory use. In this paper, we propose an explainable framework for Ethereum fraud detection integrating an XGBoost ensemble with TreeSHAP. This system achieves high predictive performance (96.3% F1-score, 96.6% recall) while providing model-level transparency via an interactive chatbot interface. Evaluation using fidelity and stability metrics confirms the reliability of the SHAP-based insights, while user-role simulations demonstrate that our structured delivery enhances clarity and actionability over standard visualizations. This work offers a practical, transparent foundation for deploying robust AI in high-risk financial environments without sacrificing accuracy.
Pedro Henrique F. S. Oliveira, Heder S. Bernardino, Saulo Moraes Villela, Edelberto Franco Silva · 6 authors
O Ethereum é uma plataforma de criptomoedas que permite a execução de contratos inteligentes, programas autônomos que operam em uma rede descentralizada. As vulnerabilidades nesses contratos representam grandes riscos financeiros e de segurança nos ecossistemas blockchain, motivando a automatização do processo de detectá-las. Este trabalho estuda a detecção de vulnerabilidades em contratos inteligentes Ethereum usando embeddings derivados de bytecode. Embeddings são representações vetoriais geradas por modelos de linguagem, que capturam as características estruturais de texto. Essas representações foram usadas como entrada para os algoritmos de regressão logística, árvore de decisão e floresta aleatória, com o fim de detectar quais contratos possuem vulnerabilidades. Os resultados mostram que os embeddings contêm informações úteis para distinguir contratos vulneráveis de não vulneráveis. O estudo também constata que a alteração da distribuição original dos dados durante o treinamento afeta significativamente o desempenho, destacando a sensibilidade das abordagens baseadas em embeddings às estratégias de amostragem.