The rapid growth of blockchain technology and the increasing number of network nodes have heightened the risk of sophisticated attacks. Among these, Eclipse attacks present a serious threat to decentralized networks by exploiting their peer-to-peer structures. While previous research has explored artificial intelligence techniques to defend against Eclipse attacks, evolving attack patterns continue to challenge existing defenses. In this paper, we propose a novel defense framework that integrates a clustering approach based on self-attention encoders within a multi-kernel neural network clustering model. Our method utilizes parallel subnetworks to extract category-specific features from multiple perspectives, generating discriminative cluster centroids that are combined with raw transaction data to train a robust classifier for detecting Eclipse attacks in Ethereum networks. To evaluate our approach, we simulate Eclipse attacks on the Ethereum testnet and conduct extensive experiments. The results demonstrate that our method achieves a detection accuracy of 98.5% and improves classification performance by 5% compared to models trained without cluster-enhanced features, confirming the effectiveness of the proposed defense.
Yan Liu, Diandong Liu, Yongyi Han, Qianying Yang · 6 authors
Smart contracts are central to decentralized applications, enabling automated execution and eliminating intermediaries. However, once deployed on the blockchain, they become immutable, making them susceptible to significant security vulnerabilities that could lead to financial losses and data breaches. Despite advancements in static analysis and symbolic execution for vulnerability detection, these methods remain limited in dynamic environments and large-scale contracts. Recently, large language models (LLMs) have shown potential in code analysis and vulnerability detection, but given concerns over the security of large models, there is an urgent need for evaluations of open-source LLMs to guide their deployment and application. This study aims to assess the performance of state-of-the-art open-source LLMs in detecting smart contract vulnerabilities. Using the Smartbugs dataset, we evaluate four prominent models-OpenAI’s gpt-oss 20B and 120B, Qwen3 235B, and DeepSeek R1 671B. The study focuses on three main areas: 1) evaluating the effectiveness of these models in detecting various vulnerabilities; 2) comparing them with existing detection tools; and 3) analyzing their limitations and potential for future development. The main contributions of this study include the first comprehensive evaluation of open-source LLMs in smart contract vulnerability detection, a detailed comparison with traditional tools, and the release of open experimental methods and data to foster further research in the field. The code used in this study can be found at https://github.com/zumAndenken/Open-Source-LLMSmartContract.
Smart contracts operating on blockchain platforms like Ethereum often face critical security challenges, which can lead to significant financial damage. Existing detection techniques frequently fall short in terms of precision or scalability. To overcome these limitations, we introduce an innovative deep learning framework that integrates Convolutional Neural Networks (CNN) with Bidirectional Long Short-Term Memory (BiLSTM) networks and incorporates an attention mechanism. This model is designed to extract both short-range syntactic cues and long-distance contextual patterns from smart contract code. By applying advanced preprocessing and employing the SMOTE algorithm to handle class imbalance, our approach excels in identifying key vulnerabilities, including Reentrancy, Timestamp Dependency, Integer Overflows, and Dangerous Delegatecall . Experimental evaluations reveal notable gains in detecting rare vulnerability types, particularly in terms of recall and precision, underscoring the effectiveness of attention-augmented hybrid models for enhancing smart contract security analysis.
AI systems rely heavily on high-quality training data, yet provenance tracking remains fragmented and vulnerable to manipulation. This study presents a blockchain-enabled data provenance framework designed to bring transparency and verifiability to AI training pipelines. The architecture records dataset lineage, preprocessing steps, annotation events and model updates using immutable smart contract transactions. A lightweight off-chain storage mechanism reduces blockchain overhead while maintaining audit guarantees. The system was tested with three machine learning pipelines involving image classification, text processing and sensor analytics. Results show a 95 percent reduction in provenance disputes and full traceability across all data contributors. Smart contracts automate compliance checks and access permissions, ensuring that only validated datasets feed into the training process. The framework improves accountability for AI ethics, model bias evaluation and regulatory reporting. Experiments confirm that blockchain latency does not significantly affect pipeline throughput due to parallelized validation nodes. The work demonstrates how decentralized technologies can support trustworthy AI development. Future research will explore integration with zero-knowledge proofs to further enhance confidentiality.
ExecMesh introduces cryptographically verifiable computation as a foundational primitive for regulatory compliance and audit trail requirements in AI/ML systems [1–3]. By combining commitmentbased verification with secure multi-party oracles and a two-tier regulatory architecture, ExecMesh enables enterprises to meet FDA, SEC, and EU AI Act requirements while maintaining the benefits of decentralized infrastructure. Immediate Value Proposition: ExecMesh provides immediate value as an audit trail and provenance layer for regulated AI systems, independent of advances in zero-knowledge proof technology. Even without full verification of large neural networks, the system delivers cryptographic guarantees for data integrity, execution timestamps, and pipeline reproducibility—meeting core regulatory requirements today.
This paper addresses the critical need for accountability in artificial intelligence (AI) systems, particularly in domains where decisions have significant societal and ethical implications. We propose a novel framework leveraging auditable attestations to ensure provable compliance with predefined standards and regulations. The core of our approach involves generating verifiable proofs about the behavior and characteristics of machine learning models, allowing for independent audits and assessments. We explore the theoretical foundations of such attestations, focusing on cryptographic techniques like zero-knowledge proofs and secure multi-party computation, which enable the verification of model properties without revealing sensitive information. Furthermore, we discuss the practical implementation of our framework, including the design of attestation protocols, the selection of relevant model properties to verify, and the development of tools for generating and validating attestations. We illustrate the effectiveness of our approach through case studies in areas such as fairness in lending, transparency in healthcare, and safety in autonomous driving. Our results demonstrate the potential of auditable attestations to enhance trust and accountability in AI systems, fostering responsible innovation and deployment.
Smart contracts are a core component of blockchain technology and are widely deployed across various decentralized scenarios. However, atomicity violations have become a critical potential security risk. Existing analysis tools often lack the precision required to detect these issues effectively. To address this challenge, we introduce AtomGraph, an automated framework designed for detecting atomicity violations. This framework leverages Graph Convolutional Networks (GCN) to accurately identify atomicity violations through multimodal feature learning and fusion. Specifically, driven by a collaborative learning mechanism, the model simultaneously learns from two heterogeneous modalities: extracting structural topological features from the bytecode-based Control Flow Graph (CFG) and uncovering deep semantics from its opcode sequence. We designed an adaptive weighted fusion mechanism to dynamically adjust the weights of features from each modality to achieve optimal feature fusion. Finally, GCN detects graph-level atomicity violations on the contract. Comprehensive experimental evaluations demonstrate that AtomGraph achieves 96.88% accuracy and 96.97% F1 score, outperforming existing tools. Furthermore, compared to the baseline concatenation fusion model, AtomGraph improves the F1 score by 6.4%, proving its potential in smart contract security detection.
Blockchain networks face a critical but understudied threat: wave attacks that exploit difficulty adjustment algorithms through strategic mining participation. Adversaries cyclically withdraw and re-enter mining to create oscillations that degrade network liveness and destabilize honest miners’ revenue. We present the first production-ready framework that maintains network responsiveness while enabling robust, post hoc threat detection. The framework employs a statistically rigorous pipeline featuring controller-aligned anomaly detection, transitive collusion grouping via union-find, and Benjamini–Hochberg False Discovery Rate control. We formally prove the economic viability of this architecture: when penalties on unvested rewards are enabled by governance, wave attacks become asymptotically unprofitable for rational adversaries. Evaluated on a 128-node distributed testbed simulating Bitcoin, Ethereum Classic, and Monacoin networks over 30 independent runs, our framework achieves 92.7% F1-score in detecting attacks, significantly outperforming baseline methods (74.7%). This work provides a complete, theoretically-grounded solution for securing proof-of-work blockchains against difficulty manipulation, forming the foundation for the adaptive AI-driven enhancements presented in our companion paper (Part II).
In recent years,the security issues of smart contracts have become increasingly prominent,and vulnerability detection has become a key challenge.In scenarios where source code is not publicly available,bytecode-based detection methods have attracted significant attention.However,existing deep learning methods typically rely solely on sequences or graph structures,which makes it difficult to fully capture vulnerability features.To address this,this paper proposes a smart contract bytecode vulnerability detection method based on heterogeneous graphs and instruction sequences,called RGCN-ResNet1D(Relational Graph Convolutional Network and ResNet-based 1D Convolutional Network).This method models bytecode as a heterogeneous graph and instruction sequence,using a Relational Graph Convolutional Network(RGCN) to extract structural features and a ResNet-based 1D Convolutional Network(ResNet1D) to extract sequential features,and then fuses the two types of features for vulnerability detection.A cross-entropy loss function is also designed,which dynamically adjusts the weight based on the number of misclassified samples,effectively alleviating the class imbalance problem in the training set.Experimental results show that RGCN-ResNet1D achieves F1 scores of 95.43%,90.67%,and 92.31% for detecting integer overflow,timestamp dependency,and self-destruct vulnerabilities,respectively,significantly outperforming the comparison methods.
Mustafa M. Abd Zaid, Husam I. Shaheen, Nigora Abduraimova, Jasim Gshayyish Zwaid · 7 authors
Smart contracts are self-executing computer-based agreements that are implemented on blockchain systems and that their security is of paramount importance because they are not subject to change. The knowledge of token-level semantics can be helpful in determining the areas that may pose a weakness in these contracts. Nevertheless, current techniques tend to be based on rule based analysis, or syntax level analysis, which find it difficult to reflect the richer semantic structures that result in complex vulnerabilities. In order to overcome these limitations, this paper presents a framework that combines the pretrained transformer functionality of CodeBERT with task-specific fine-tuning and, as such, auto-detects and highlights vulnerabilities in smart contract Integrated Development Environment (IDEs). The method is an examination of token-level semantics, making it possible to identify vulnerabilities and understand them correctly in context. This framework can be applied directly in real-time to IDEs by developers to get vulnerability notifications and recommendations. The experimental outcomes prove that FTC-BERT is much more effective in detecting vulnerabilities and remembering experiments than traditional, and it is a semantic-sensitive, efficient, and automated method to detect vulnerabilities in smart contracts.
This paper presents LLMBugScanner, a large language model (LLM) based framework for smart contract vulnerability detection using fine-tuning and ensemble learning. Smart contract auditing presents several challenges for LLMs: different pretrained models exhibit varying reasoning abilities, and no single model performs consistently well across all vulnerability types or contract structures. These limitations persist even after fine-tuning individual LLMs. To address these challenges, LLMBugScanner combines domain knowledge adaptation with ensemble reasoning to improve robustness and generalization. Through domain knowledge adaptation, we fine-tune LLMs on complementary datasets to capture both general code semantics and instruction-guided vulnerability reasoning, using parameter-efficient tuning to reduce computational cost. Through ensemble reasoning, we leverage the complementary strengths of multiple LLMs and apply a consensus-based conflict resolution strategy to produce more reliable vulnerability assessments. We conduct extensive experiments across multiple popular LLMs and compare LLMBugScanner with both pretrained and fine-tuned individual models. Results show that LLMBugScanner achieves consistent accuracy improvements and stronger generalization, demonstrating that it provides a principled, cost-effective, and extensible framework for smart contract auditing.
P. Prakash, Faheema Kattakath Sanil, Jeffrey Tom Shaji, Saravanan Palani · 5 authors
The adoption of privacy-preserving techniques in healthcare is significant, especially while handling sensitive medical information. Traditional machine learning approaches raise significant concern regarding privacy, regulations, and data protection. Federated learning has emerged as an effective machine learning approach that enables a group of local models to collaboratively train the global model by sharing their updates instead of sharing the sensitive medical data. Nevertheless, a significant issue with federated learning is its vulnerability to various attacks, including model corruption and data tampering. The authors propose a methodology for developing a secure and privacy-safeguarded collaborative learning model by integrating zero knowledge proof (ZKP) with federated learning (FL). The proposed RS-ZKP methodology utilizes Pedersen commitments within ZKP to verify feature importance, ensuring that they fall within specified bounds without disclosing the actual values. The methodology is validated on two benchmark datasets using metrics accuracy, precision, recall, and F1 score.
One of the most promising technology of blockchain still suffers from multiple attacks. Attack detection is handled by various technologies to enhance the security. Hackers moto is to steal the digital currency here the discussion is of the digital currency Ethereum which also has suffered from the famous attack Re-entrancy attack. Though the usage of smart contact is highly believed by users, making the function to be called again and again leads to this kind of attack. Re-entrancy attacks pose a persistent threat to Ethereum smart contracts, enabling malicious users to repeatedly call vulnerable functions before the state variables are correctly updated. While traditional approaches rely heavily on symbolic execution and SMT solvers for vulnerability detection, these methods suffer from scalability and precision limitations. A novel hybrid framework that integrates dynamic execution tracing with AI-driven classification models, such as Quantum Neural Networks (QNN), for the robust detection of re-entrancy attacks. The system leverages taint analysis and runtime monitoring to generate behavioral features, which are then used to train and evaluate machine learning models. Experiments conducted on benchmark datasets demonstrate superior accuracy and reduced false positives when compared to conventional static analysis techniques, showcasing the effectiveness of combining dynamic and AI-based strategies for smart contract security.
Jiaxi Liu, Lin Sun, Tianyu Kang, Di Wu · 7 authors
Federated Learning (FL) enables model training on distributed devices while preserving data privacy. However, malicious clients can submit fabricated model updates to fraudulently obtain training rewards, a behavior known as free-rider attacks. Existing detection-based solutions analyze anomalies in model updates but lack direct evidence of local training, making it fail to fully prevent free-riders. To address this limitation, we propose zkVFL, a verifiable FL framework leveraging Zero-Knowledge Proofs (ZKP) to ensure the integrity of local training while preserving privacy. To reduce the computational overhead of proof generation in ZKP, zkVFL introduces two novel techniques: (i) anomaly-aware client sampling to selectively perform ZKP verification and (ii) A recursive ZKP protocol (ReMPoT), incorporating a pruning-based layer selection technique, reduces proof generation costs. Experimental results demonstrate that zkVFL improves the accuracy and convergence of FL training under free-rider attacks while significantly reducing the computational and memory overhead of proof generation on resource-constrained devices.
This paper unveils a pioneering modular framework for Decentralized Explainable Artificial Intelligence (DeXAI), harnessing blockchain to deliver unparalleled trust and clarity in AI systems. Addressing the opacity and privacy challenges of centralized AI, our framework integrates federated learning with Explainable AI (XAI) methods, namely SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME), to produce intuitive explanations for AI decisions across distributed networks. A blockchain layer secures predictions and explanations, with smart contracts ensuring ethical compliance and auditable trails. Designed for adaptability, the framework supports diverse AI models and blockchain platforms, excelling in critical sectors like healthcare and finance. Our prototype validates its scalability and effectiveness, setting a new benchmark for trustworthy AI.
Muhammad Sannan Khaliq, Sium Bin Noor, Subroto Kumar Ghosh, Love Allen Chijioke Ahakonye · 6 authors
Smart contracts are integral to blockchain applications; yet, their immutability creates security vulnerabilities, such as reentrancy and overflow, which can be critically damaging. While detection tools exist, many rely on symbolic execution, graph preprocessing, or binary classification, limiting their efficiency and practicality. This study presents an optimized DeBERTa V3-based transformer model for multilabel detection of smart contract vulnerabilities. The proposed approach operates directly on tokenized Solidity code, leveraging disentangled attention and position embeddings to model semantic patterns. Evaluation on three public datasets achieves up to 100% F1-scores on key vulnerabilities and maintains an average inference latency below 58 ms per smart contract. These results demonstrate the feasibility of integrating high-accuracy, low-latency vulnerability detection into real-time auditing tools, thereby enhancing contract security before deployment.
Recent advances in artificial intelligence (AI), particularly deep learning, have led to widespread adoption across various applications. Yet, a fundamental challenge persists: how can we verify the correctness of AI model inference when model owners cannot (or will not) reveal their parameters? These parameters represent enormous training costs and valuable intellectual property, making transparent verification difficult. In this paper, we introduce a zero-knowledge framework capable of verifying deep learning inference without exposing model internal parameters. Built on recursively composed zero-knowledge proofs and requiring no trusted setup, our framework supports both linear and nonlinear neural network layers, including matrix multiplication, normalization, softmax, and SiLU. Leveraging the Fiat-Shamir heuristic, we obtain a succinct non-interactive argument of knowledge (zkSNARK) with constant-size proofs. To demonstrate the practicality of our approach, we translate the DeepSeek model into a fully SNARK-verifiable version named ZK-DeepSeek and show experimentally that our framework delivers both efficiency and flexibility in real-world AI verification workloads.
Modern cyber threats, known for their complexity and constant change, surpass traditional intrusion detection systems (IDS). This paper explores a new security approach that combines Artificial Intelligence (AI) with decentralized architectures to develop IDS that are robust, scalable, and protect user privacy. It examines the core roles of Federated Learning (FL) and Blockchain, highlighting three main research challenges: The vulnerability of AI models to adversarial attacks, privacy and data integrity concerns in collaborative learning, and performance limitations in distributed systems. To address these issues, we suggest solutions such as adversarial training, differential privacy, and lightweight consensus mechanisms. Our analysis of case studies shows that hybrid FL-Blockchain systems outperform traditional methods in practical application environments.
Blockchain smart contracts have been a groundbreaking technology yet are still prone to numerous security vulnerabilities that can lead to large financial and operational losses. In our prior work, we presented an in-depth methodology for data preprocessing and dataset preparation towards facilitating effective vulnerability detection in smart contracts. In this paper, building on that work, we proceed with our work by presenting a new hybrid deep learning architecture that integrates Graph Neural Networks (GNN) and CodeBERT in an efficient way to capture both structural and semantic code features. The hybrid model processes parallel representations of smart contracts: the GNN extracts graph-based control and data flow dependency relations, and CodeBERT makes use of pretrained contextual source code token embeddings. The two embeddings are concatenated and then fed into a shared classifier to predict the existence and types of vulnerabilities. We test our approach on a diverse collection of smart contracts and compare it with single-model baselines. Our hybrid model outperforms individual GNN and CodeBERT approaches with significant performance gains in precision, recall, and F1-score for different types of vulnerabilities. These findings confirm the effectiveness of our fusion approach and introduce the possibility of employing hybrid deep learning models in real-world smart contract security auditing.
The increasing use of deep learning (DL) models has given rise to significant privacy concerns regarding training and inference data. To address these concerns, the community has increasingly adopted crypto-based privacy-enhancing technologies (CPET) like homomorphic encryption (HE), secure multi-party computation (MPC), and zero-knowledge proofs (ZKP). The integration of CPET with DL, often referred to as CPET-DL, is commonly facilitated by specialized frameworks like CrypTen, TenSEAL, and EZKL. These frameworks offer configurable parameters to balance model accuracy and computational efficiency during privacy-preserving operations. However, these configurations, while seemingly harmless, can introduce subtle vulnerabilities. The stealthy attacks induced by misconfigurations are hard to detect because 1) the plaintext models remain vulnerability-free, and 2) existing auditing tools are hardly applicable to CPET-hardened models. This creates a paradox: tools intended to protect privacy can be undermined through configuration manipulation.
Open access
Cryptography and Data Security
Physical Unclonable Functions (PUFs) and Hardware Security
M. Saravana Karthikeyan, R. Rajasree, R. Santhana Krishnan, C. Gayathri · 6 authors
Secure and efficient healthcare data sharing is critical for modern medical ecosystems, yet existing systems often suffer from limited scalability, privacy risks, and lack of intelligent data management. This study proposes MedVault, a hybrid blockchain-cloud-AI framework designed for secure, patient-centric healthcare data management. The architecture employs Corda for on-chain storage of consent records, metadata, and audit logs, while large medical datasets are encrypted and stored off-chain in AWS S3 with PostgreSQL metadata management. Security is reinforced using AES-256 encryption, Proxy Re-Encryption (PRE), Zero-Knowledge Proofs (ZKP), and decentralized identity management via Hyperledger Indy and Aries, while a FHIR-based gateway ensures seamless integration with electronic health records (EHRs). Intelligence is incorporated through deep learning models, including Autoencoders for anomaly detection, CNNLSTM for medical data analytics, Graph Neural Networks (GNNs) for consent prediction, DNNs for risk assessment, and Federated Learning (FL) for privacy-preserving distributed model training. Variational Autoencoders (VAEs) generate synthetic datasets, and Explainable AI techniques (SHAP, LIME) ensure interpretability. Extensive evaluations demonstrate that Corda-MedVault outperforms Hyperledger Fabric, Ethereum, and traditional centralized approaches across metrics such as blockchain latency, throughput, auditability, off-chain storage efficiency, energy consumption, anomaly detection, and consent prediction. Overall, the proposed system provides a scalable, energy-efficient, privacypreserving, and intelligent platform for real-time healthcare data sharing, offering a robust solution for secure and compliant medical data management.