• Introduces a workflow-oriented taxonomy for backdoor attacks in FL. • A bi-dimensional taxonomy classifies defenses by mechanism and FL stage. • Empirically evaluates the effectiveness of SoTA attacks and defenses. • Identifies key limitations and outlines future research directions. Federated Learning (FL) is a collaborative paradigm that enables decentralized model training without centralizing data. However, FL remains highly vulnerable to backdoor attacks, where adversaries poison local updates to insert hidden malicious behaviors into the global model. Over the past few years, a rapidly growing body of work has proposed both attack strategies and defense mechanisms, yet the field remains chaotic, with inconsistent assumptions, evaluation practices, and a lack of clear understanding of the core trade-offs. In this paper, we present a comprehensive Systematization of Knowledge (SoK) of the field. We introduce novel, multi-dimensional taxonomies to deconstruct attacks and categorize defenses by their intervention point and underlying techniques. Ultimately, our analysis reveals a critical gap between research and practice, highlighting unaddressed challenges in scalability, data heterogeneity, and the conflict between privacy and robustness.

International audience

Roaming settlement in 5G and beyond networks demands secure, efficient, and trustworthy mechanisms for billing reconciliation between mobile operators. While blockchain promises decentralization and auditability, existing solutions suffer from critical limitations—namely, data privacy risks, assumptions of mutual trust, and scalability bottlenecks. To address these challenges, we present B5GRoam, a novel on-chain and zero-trust framework for secure, privacy-preserving, and scalable roaming settlements. B5GRoam introduces a cryptographically verifiable call detail record (CDR) submission protocol, enabling smart contracts to authenticate usage claims without exposing sensitive data. To preserve privacy, we integrate non-interactive zero-knowledge proofs (zkSNARKs) that allow on-chain verification of roaming activity without revealing user or network details. To meet the high-throughput demands of 5G environments, B5GRoam leverages Layer 2 zk-Rollups, significantly reducing gas costs while maintaining the security guarantees of Layer 1. Experimental results demonstrate a throughput of over 7,200 tx/s with strong privacy and substantial cost savings. By eliminating intermediaries and enhancing verifiability, B5GRoam offers a practical and secure foundation for decentralized roaming in future mobile networks.

The rapid growth of decentralized finance (DeFi) has led to the widespread use of automated agents, or bots, within blockchain ecosystems like Ethereum, Binance Smart Chain, and Solana. While these bots enhance market efficiency and liquidity, they also raise concerns due to exploitative behaviors that threaten network integrity and user trust. This paper presents a decentralized federated learning (DFL) approach for detecting financial bots within Ethereum Virtual Machine (EVM)-based blockchains. The proposed framework leverages federated learning, orchestrated through smart contracts, to detect malicious bot behavior while preserving data privacy and aligning with the decentralized nature of blockchain networks. Addressing the limitations of both centralized and rule-based approaches, our system enables each participating node to train local models on transaction history and smart contract interaction data, followed by on-chain aggregation of model updates through a permissioned consensus mechanism. This design allows the model to capture complex and evolving bot behaviors without requiring direct data sharing between nodes. Experimental results demonstrate that our DFL framework achieves high detection accuracy while maintaining scalability and robustness, providing an effective solution for bot detection across distributed blockchain networks.

The Internet of Electric Vehicles (IoEV) envisions a tightly coupled ecosystem of electric vehicles (EVs), charging infrastructure, and grid services, yet remains vulnerable to cyberattacks, unreliable battery-state predictions, and opaque decision processes that erode trust and performance. To address these challenges, we introduce a novel Agentic Artificial Intelligence (AAI) framework tailored for IoEV, where specialized agents collaborate to deliver autonomous threat mitigation, robust analytics, and interpretable decision support. Specifically, we design an AAI architecture comprising dedicated agents for cyber-threat detection and response at charging stations, real-time State of Charge (SoC) estimation, and State of Health (SoH) anomaly detection, all coordinated through a shared, explainable reasoning layer; develop interpretable threat-mitigation mechanisms that proactively identify and neutralize attacks on both physical charging points and learning components; propose resilient SoC and SoH models that leverage continuous and adversarial-aware learning to produce accurate, uncertainty-aware forecasts with human-readable explanations; and implement a three-agent pipeline, where each agent uses LLM-driven reasoning and dynamic tool invocation to interpret intent, contextualize tasks, and execute formal optimizations for user-centric assistance. Finally, we validate our framework through comprehensive experiments across diverse IoEV scenarios, demonstrating significant improvements in security and prediction accuracy. All datasets, models, and code will be released publicly.

Blockchain-based Federated Learning (BFL) is an emerging decentralized machine learning paradigm that enables model training without relying on a central server. Although some BFL frameworks are considered privacy-preserving, they are still vulnerable to various attacks, including inference and model poisoning. Additionally, most of these solutions employ strong trust assumptions among all participating entities or introduce incentive mechanisms to encourage collaboration, making them susceptible to multiple security flaws. This work presents VerifBFL, a trustless, privacy-preserving, and verifiable federated learning framework that integrates blockchain technology and cryptographic protocols. By employing zero-knowledge Succinct Non-Interactive Argument of Knowledge (zk-SNARKs) and in-crementally verifiable computation (IVC), VerifBFL ensures the verifiability of both local training and aggregation processes. The proofs of training accuracy and aggregation are verified on-chain, guaranteeing the integrity and auditability of each participant's contributions. To protect training data from inference attacks, VerifBFL leverages differential privacy. Finally, to demonstrate the efficiency of the proposed protocols, we built a proof of concept using emerging tools. The results show that generating proofs for local training and aggregation in VerifBFL takes less than 81s and 2s, respectively, while verifying them on-chain takes less than 0.6s.

Blockchained federated learning (BFL) combines the concepts of federated learning and blockchain technology to enhance privacy, security, and transparency in collaborative machine learning models. However, implementing BFL frameworks poses challenges in terms of scalability and cost-effectiveness. Reputation-aware BFL poses even more challenges, as blockchain validators are tasked with processing federated learning transactions along with the transactions that evaluate FL tasks and aggregate reputations. This leads to faster blockchain congestion and performance degradation. To improve BFL efficiency while increasing scalability and reducing on-chain reputation management costs, this paper proposes AutoDFL, a scalable and automated reputation-aware decentralized federated learning framework. AutoDFL leverages zk-Rollups as a Layer-2 scaling solution to boost the performance while maintaining the same level of security as the underlying Layer-1 blockchain. Moreover, AutoDFL introduces an automated and fair reputation model designed to incentivize federated learning actors. We develop a proof of concept for our framework for an accurate evaluation. Tested with various custom workloads, AutoDFL reaches an average throughput of over 3000 TPS with a gas reduction of up to 20X.

Artificial intelligence (AI) is increasingly used in the automotive industry for applications such as driving style classification, which aims to improve road safety, efficiency, and personalize user experiences. While deep learning (DL) models, such as Long Short-Term Memory (LSTM) networks, excel at this task, their "black-box" nature limits interpretability and trust. This paper proposes a machine learning (ML)-based method that balances high accuracy with interpretability. We introduce a high-quality dataset, "CARLA-Drive", and leverage ML techniques like Random Forest (RF), Gradient Boosting (XGBoost), and Support Vector Machine (SVM), which are efficient, lightweight, and interpretable. In addition, we apply the SHAP (Shapley Additive Explanations) explainability technique to provide personalized recommendations for safer driving. Achieving an accuracy of 0.92 on a three-class classification task with both RF and XGBoost classifiers, our approach matches DL models in performance while offering transparency and practicality for real-world deployment in intelligent transportation systems.

As the Internet of Things (IoT) continues to expand, botnet-driven threats pose a growing and severe risk to the security of IoT-enabled infrastructures. These threats exploit large numbers of compromised devices to establish covert control channels and, eventually, launch large-scale cyberattacks such as Distributed Denial of Service (DDoS), capable of severely disrupting critical services and causing substantial economic damage. This paper highlights the urgent need for detecting botnets at an early stage, particularly by identifying stealthy command and control (C&C) traffic that precedes the execution of such attacks. We propose an anomaly-based detection framework that combines semi-supervised learning with explainable Artificial Intelligence (XAI). Unlike most existing approaches, our method requires only benign traffic for training, thereby enabling the detection of previously unseen or evolving botnet threats without relying on labeled malicious data. The framework supports multiple traffic representations, including raw bytes, packet-level data, and unidirectional or bidirectional flows, enriched with diverse network features to enhance detection coverage and adaptability. Experimental evaluations using the IoT-23 dataset demonstrate a 99.51% detection rate and a 1.09% false positive rate for stealthy C&C communications, underscoring the method’s effectiveness and robustness. The integration of XAI enhances transparency and interpretability, enabling security professionals to better understand model decisions and refine detection strategies.

Current blockchain-based reputation solutions for crowdsourcing fail to tackle the challenge of ensuring both efficiency and privacy without compromising the scalability of the block chain. Developing an effective, transparent, and privacy-preserving reputation model necessitates on-chain implementation using smart contracts. However, managing task evaluation and reputation updates alongside crowdsourcing transactions on-chain substantially strains system scalability and performance. This paper introduces RollupTheCrowd, a novel blockchain-powered crowdsourcing framework that leverages zkRollups to enhance system scalability while protecting user privacy. Our framework includes an effective and privacy-preserving reputation model that gauges workers' trustworthiness by assessing their crowdsourcing interactions. To alleviate the load on our blockchain, we employ an off-chain storage scheme, optimizing RollupTheCrowd's performance. Utilizing smart contracts and zero-knowledge proofs, our Rollup layer achieves a significant 20x reduction in gas consumption. To prove the feasibility of the proposed framework, we developed a proof-of-concept implementation using cutting-edge tools. The experimental results presented in this paper demonstrate the effectiveness and scalability of RollupTheCrowd, validating its potential for real-world application scenarios.

Large Language Models (LLMs) have witnessed a rapid growth in emerging challenges and capabilities of language understanding, generation, and reasoning. Despite their remarkable performance in natural language processing-based applications, LLMs are susceptible to undesirable and erratic behaviors, including hallucinations, unreliable reasoning, and the generation of harmful content. These flawed behaviors under-mine trust in LLMs and pose significant hurdles to their adoption in real-world applications, such as legal assistance and medical diagnosis, where precision, reliability, and ethical considerations are paramount. These could also lead to user dissatisfaction, which is currently inadequately assessed and captured. Therefore, to effectively and transparently assess users' satisfaction and trust in their interactions with LLMs, we design and develop LLMChain, a decentralized blockchain-based reputation system that combines automatic evaluation with human feedback to assign contextual reputation scores that accurately reflect LLM's behavior. LLMChain helps users and entities identify the most trustworthy LLM for their specific needs and provides LLM developers with valuable information to refine and improve their models. To our knowledge, this is the first time that a blockchain-based distributed framework for sharing and evaluating LLMs has been introduced. Implemented using emerging tools, LLMChain is evaluated across two benchmark datasets, showcasing its effectiveness and scalability in assessing seven different LLMs.