Authenticated Private Information Retrieval (Authenticated PIR) allows the client to retrieve the desired database entry without revealing any information about the query, while safely aborting if malicious behavior by the server is detected (presented in USENIX ’23). However, two key challenges remain: existing single-server authenticated PIR schemes with sublinear online communication have not yet been clearly and fully implemented; incremental updates to the digest introduce unnecessary overhead. In this paper, we implement the previously outlined idea and present two complete and concrete single-server authenticated PIR schemes with O(√(N)) online communication, namely LWE-AuthPIR and DDH-AuthPIR, along with detailed security proofs. Furthermore, we introduce the notion of single-server authenticated and incremental PIR and propose a corresponding concrete construction, LWE-AuthIncPIR. LWE-AuthIncPIR supports immediate updates to individual entries and integrates a communication-efficient row aggregation for periodic update scenarios. When 1 × compared to previous incremental update method. In the password breach detection, LWE-AuthIncPIR achieves a 67 × reduction in preprocessing time and a 2.9 × reduction in communication overhead.
With the continual evolution of network-attack techniques, conventional password-based authentication mechanisms have become increasingly ineffective when countering emerging cyber threats. To improve both security and usability, the FIDO Alliance introduced the FIDO2 authentication standard. As a passwordless solution, FIDO2 leverages public-key cryptography to mitigate risks such as credential leakage and to provide a more secure means of identity verification. Nevertheless, real-world FIDO2 deployments can still be exposed to latent security threats. To address this gap, we propose a FIDO2 security assessment framework grounded in a Markov Decision Process (MDP). By modelling the FIDO2 authentication flow as an MDP, the framework evaluates the in-transit security risk of WebAuthn communications and their associated verification parameters. It dynamically calculates the risk state of each authentication request in real time, adaptively escalates the strength of the FIDO2 authentication mechanism, and thus reinforces the transmission security of FIDO2-based identity verification.
We propose a self-supervised learning framework for identifying secondary compression artifacts in H.264 videos, which tackles the challenge of scarce labeled data in forensic analysis. The proposed method includes a Compression Artifact Prediction Module (CAPM) that trains a convolutional neural network (CNN) beforehand to estimate synthetic compression parameters from unlabeled video frames, which helps the model acquire discriminative features without requiring extensive annotations. The system produces synthetic artifacts by modeling controlled re-compression scenarios with different quantization parameters and frame rates, which act as self-supervision signals for pre-training. The CAPM employs a multi-task loss to jointly estimate compression parameters, frame type distributions, and motion vector magnitudes, thereby capturing intrinsic patterns of compression artifacts. Additionally, a module dedicated to temporal consistency combines spatiotemporal features from multiple frames by applying 3D convolutions, whereas a contrastive detection head improves the distinction between primary and secondary compression traces. The last detection phase merges transformer-driven self-attention and a binary classifier to capture extended dependencies and strengthen decision reliability. Experiments show that our method attains good performance in secondary compression detection while markedly decreasing dependence on annotated data.
Cross-chain asset bridges usually require the joint authorization of multiple nodes to execute transactions. Multi-signature, due to its ability to compress multiple signatures for the same message into a short signature, can be well applied to cross-chain asset bridges. The existing multi-signature schemes, however, based on distributed key generation (DKG) have obviously three vulnerabilities: the key generation stage incurs considerable interactive communications, a malicious centralized combiner enables to tamper with transaction content, and the signature scheme lacks the accountability function for signers. To solve these challenging issues, this paper constructs a new type of traceable multi-signature scheme TMDC (Traceable Multi-signature Scheme with Distributed Combiners). In this scheme, a fully decentralized key generation is exploited, where the signer node can independently generate public and private key pairs. To avoid the centralization of combiners, time-lock is used to limit the aggregation power of combiners. Once the time lock is unlocked, the combiner can no longer aggregate. Furthermore, TMDC designs the signature accountability algorithm and establishes a strong correlation mechanism between the signer’s behavior and the final multi-signature. Based on the specific properties above-mentioned, TMDC can be well applied to cross-chain asset bridges. TMDC not only enables distributed combiners to avoid the risks brought by only one combiner, but also can trace abnormal signers when the cross-chain asset bridge is attacked, significantly enhancing the security of the cross-chain asset bridge. Finally, after simultaneously implementing functions, TMDC still maintains a low communication and computational overhead. The verifying algorithm is superior to other related schemes, significantly enhancing the system efficiency.
Adversarial attacks induce misclassification in deep neural networks through imperceptible perturbations to human vision. Traffic sign images exhibit intrinsic structural vulnerabilities in decision boundaries due to their characteristic features—high-saturation color distributions and standardized geometric shapes. To address these vulnerabilities, this paper proposes a clean-sample guided adaptive overlapping patch defense method specifically designed for traffic sign recognition systems. Diverging from conventional perturbation-elimination paradigms, our approach constructs defense mechanism by exploiting the visual indistinguishability between adversarial and clean samples. A clean sample repository is established through K-means and Histogram of Oriented Gradients feature modeling, with rapid input-to-repository matching achieved via dual metrics of “structural similarity” and “perceptual hashing” joint similarity. Guided defense patches are dynamically generated based on the nearest clean samples. Experimental results on the GTSRB dataset show that our method maintains an original classification accuracy of 0.9838, merely 0.01 lower than the baseline model, while achieving a defensive accuracy of 0.7240 against PGD attacks, comparable to current state-of-the-art defense methods. Compared to the average 0.19 accuracy degradation of traditional methods, our approach effectively mitigates the trade-off conflict between model robustness and classification accuracy.
The automotive industry is rapidly moving toward new energy, intelligence, and connectivity, driven by policies, technological advances, and growing environmental awareness, which are pushing more people to adopt electric vehicles (EVs). Moreover, EVs are driven as distributed energy storage units by Vehicle-to-Grid (V2G), which facilitates a two-way exchange of energy. The grid and vehicles can change the power. By managing when and how EVs charge, V2G helps organize the charging process, making EVs more flexible and valuable energy resources. However, a major challenge in V2G is the authentication between EVs and charging stations (CSs), as attackers may impersonate legitimate EVs to gain unauthorized access, wasting resources and harming real users. Current authentication methods also suffer from weaknesses in key security and user privacy. In this paper, a novel method based on Cryptographic Reverse Firewalls (CRFs) is proposed, which uses random numbers to ensure the security of the private key and prevent leakage and malicious tampering. This makes it difficult for attackers to steal original keys. Moreover, the proposed scheme can conceal the relationship between signatures and vehicle owners, thereby ensuring user anonymity. Security analysis shows that the proposed scheme can resist spoofing attacks, prevent signature forgery, and ensure secure key management. Besides, experimental results confirm that the scheme is efficient, making it suitable for practical V2G applications.
Network data packet anomaly detection faces numerous challenges, including incomplete supervision signals, extremely imbalanced data, weak adaptability in across- scenario. This paper proposes a GPT-based causal language model for network intrusion detection, which called NIDS-GPT. Unlike previous work, NIDS-GPT innovatively treats each number in the packet as an independent “word” rather than packet fields, enabling a more fine-grained data representation. In this model, we design special tokenizers and embedding layers to capture the structure and semantics of network data capture the structure and semantics of network data better. NIDS-GPT has good scalability and supports unsupervised pre-training. Experiments on the CICIDS2017 and car-hacking datasets show that NIDS-GPT achieves high accuracy under extreme imbalance conditions, far surpassing traditional methods; it also achieves over 90
The goal of video steganography is to embed secret information into video content in a way that it remains imperceptible, while ensuring that the embedded information can be reliably extracted by the receiver. Owing to their large data capacity and rich temporal dynamics, videos offer strong potential for covert information embedding. However, the complex structure of video data and the high compression imposed by modern video encoding algorithms pose significant challenges. Existing deep learning-based video steganography frameworks often struggle to maintain robustness under such compression. In this work, we propose an end-to-end video steganography model based on optical flow guidance and discrete wavelet transform (DWT). The model supports high-capacity embedding while preserving visual quality under video compression and transcoding. It consists of an encoder and a decoder, where the encoder leverages optical flow maps to capture temporal motion information and performs fine-grained embedding in the frequency domain via DWT. A distortion simulation layer is integrated into the network to mimic real-world degradation scenarios such as video compression attacks. Experimental results demonstrate that the proposed method achieves high embedding capacity and accurate information retrieval, while maintaining superior visual quality. Moreover, it exhibits strong robustness against video compression, highlighting its potential for practical steganographic applications.
The Border Gateway Protocol faces persistent risks from route hijacking, mitigated by Resource Public Key Infrastructure (RPKI) through cryptographically validated Route Origin Authorizations. However, existing RPKI validators need a formally verified reference for correctness, especially resource certificate validations. This paper design and present a formally verified RPKI validator to validate resource certificates. Built upon ARMOR, a formally verified X.509 validator for Transport Layer Security, our validator support parsing resource extensions and formalizing 17 semantic rules for RPKI certificate validations in RFC 6487. The validator can not only serve as a reference for testing existing RPKI validator implementations, but also be embedded into them to enhance their formal verification attributes.
Ascon was selected as the standard for lightweight cryptography by NIST in 2023. At EUROCRYPT 2023, Qin et al. introduced the meet-in-the-middle (MitM) preimage attack with a single match on sponge-based hashing. We further propose a Multi-Match technique to reduce the time complexity of the MitM preimage attack. When the neutral sets of the initial state are fixed, different MitM characteristics correspond to different matches. Multi-Match technique is the use of different matches to filter all values of the fixed neutral sets. For each candidate pair that satisfies the multi matches, a full-state match is performed. Using the Multi-Match technique, we present the improved preimage attacks on 3-round and 4-round Ascon-XOF.
The Fiat-Shamir with Aborts (FSwA) paradigm is a well-established approach to construct lattice-based signatures, leading to many efficient and practical schemes. One of popular implementations is Dilithium, which has been selected as the standardization in the post-quantum project by NIST. An important question now is whether the signature schemes following the FSwA framework allow for additional features, such as aggregation of different signatures. To our knowledge, only a few schemes have been proposed, most of them lack a sound security proof or suffer from inefficiency, highlighting the need for further research. Motivated by this circumstance, we propose a FSwA-based non-interactive aggregate signature scheme which not only achieves the aggregate signature smaller than the trivial concatenation of all individual signatures, but also gives a provable security proof. In addition, we review the state of the art of the FSwA-based aggregate signatures, including the schemes presented by Boudgoust and Roux-Langlois in the Computer Journal (2023) and by Tomita and Shikata (IACR eprint 2023/471), and observe certain issues in their constructions which inspire our design.
With the advancement of Artificial Intelligence and big data, federated learning (FL) has demonstrated significant application value. However, traditional aggregation node selection methods in FL lack the ability to adjust dynamically. Furthermore, node reputation scores are susceptible to external interference, affecting training performance, slowing model convergence, and weakening the system’s ability to handle complex environments and risks. Hence, this paper proposes a dynamic aggregation node selection scheme implemented using reputation scoring. To address the challenges of dynamic adjustment and comprehensive evaluation in node selection, this method incorporates six evaluation dimensions, namely, direct and objective reputation scoring, enabling dynamic and optimized node selection. To enhance system security and stability, cryptographic algorithms, signature verification techniques, and blockchain technologies are integrated to ensure data privacy and system integrity. The security analysis demonstrates that the proposed scheme enables the FL system to operate efficiently and securely, effectively resists malicious nodes, and improves the reliability of the system.
S-box is the main nonlinear component of block ciphers. In order to efficiently construct 8-bit S-boxes with excellent cryptographic properties, a novel 8-bit S-box construction technique based on the improved genetic algorithm is proposed. Firstly, a mixed population samples is generated to improve the nature of the initial population. Secondly, a segmented fitness function, taking into account differential uniformity, nonlinearity, and the strict avalanche criterion, is designed to evaluate individual candidate population. Note that the bijectivity of S-box populations can be ensured during the crossover operation. During the mutation operation, a reverse search method based on the properties of the linear approximation table (LAT) and the differential distribution table (DDT) of an S-box is proposed, and a 2 N-round searching strategy, combined with the hill-climbing algorithm, the random search and the reverse search technique, is designed to search mutation points to optimize the cryptographic properties of the constructed S-boxes. Additionally, multi-threading technology is used for parallel computing to optimize the computational efficiency. Experimental results demonstrate that the proposed technique can construct S-boxes with excellent cryptographic properties efficiently, achieving a differential uniformity of 4, a nonlinearity of 112, while also satisfying the bijectivity property.
Outsourced computation requires both data confidentiality and computation integrity. While fully homomorphic encryption (FHE) ensures strong confidentiality guarantees, it suffers from substantial computational and storage overhead. Homomorphic secret sharing (HSS) has emerged as a more efficient alternative to FHE, enabling multiple non-interacting servers to locally evaluate a function on their shares. To ensure computation integrity, verifiable homomorphic secret sharing (VHSS) has been proposed. However, most existing VHSS schemes support only private verification, where only the data owner can verify the result. This limits their applicability in scenarios where the result must be trusted by external parties. Publicly verifiable HSS (PVHSS) addresses this limitation by enabling anyone to verify the computation correctness with a public key. However, existing PVHSS schemes for polynomial functions incur high verification overhead. In this work, we propose a new PVHSS scheme for polynomial functions, building upon the HSS scheme of Dao et al. (CRYPTO’23). Our scheme enables any third party to verify whether multiple non-interacting servers have correctly evaluated the function. Moreover, our construction ensures that any set of up to t servers learn nothing about the outsourced data and cannot generate an incorrect computation result. Compared to existing PVHSS schemes for polynomials, our approach achieves smaller share sizes and lower verification costs.
The heterogeneity of client-side data in federated recommender systems poses challenges in training a unified model that can cater to the individual needs of each user for recommendation performance. To address this issue, we propose a recommendation algorithm for personalized federated matrix decomposition called HE-MD-PFedMF (Personalized Federated Matrix Factorization Recommendation Algorithm Based on Homomorphic Encryption and Meta-Distillation). HE-MD-PFedMF incorporates user bias item parameters, project bias term parameters, and score record average parameters into the model training process, thereby accommodating variations in project scoring standards among different users. By integrating meta-learning and knowledge distillation techniques into the federated matrix factorization model, our algorithm aims to enhance personalized recommendation accuracy by leveraging user terminal preference data comprehensively. Additionally, we employ homomorphic encryption technology to encrypt gradient parameters, mitigating concerns regarding privacy leakage during the transmission of model parameters. Experimental analysis demonstrates that HE-MD-PFedMF yields lower Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) values compared to prominent baseline algorithms, thereby achieving superior recommendation accuracy.
The present Ejar platform for real estate contract management encounters problems with trust, security, and authenticity of records, which result in disputes and fraudulent activities. This work introduces a Blockchain-Based Real Estate Contract Authentication System to address these issues through tamper-proof, transparent, and verifiable contract management. The system integrates blockchain’s immutable features with decentralized consensus mechanisms into Ejar to enable secure contract registration, validation, and retrieval. The proposed system is built on smart contracts and includes property ID-based search functionality and graphical user interfaces (GUIs) for property management and transaction confirmation. The implementation utilizes a private Ethereum blockchain, Solidity smart contracts, and a web-based user interface. Experimental results demonstrate that the system prevents unauthorized modifications, enables rapid verification of property and contract information, and supports efficient search and retrieval operations, thereby enhancing trust and operational efficiency in real estate transactions.
Fileless malware presents a growing cybersecurity challenge due to its stealthy, memory-resident behavior that evades traditional detection methods. In this paper, we propose G-FLEX (Graph-based Fileless maLwarE eXplanation), a novel framework that combines graph representations, fine-tuned transformers, and explainable AI (XAI) to detect and interpret fileless malware threats. Our method begins with the transformation of source code into Abstract Syntax Trees (ASTs) and Control Flow Graphs (CFGs) to structure code behavior and eliminate irrelevant noise. To capture deep semantic and syntactic features, we fine-tune state-of-the-art language models such as BERT, Electra, and CodeBERT, and incorporate Hierarchical Transformers to model multi-level code dependencies. GraphCodeBERT is further employed to generate enriched graph-aware embeddings. For interpretability, we apply XAI techniques, notably SHAP, to identify influential features such as function calls, AST node patterns, and CFG structures that drive the model’s decisions. Experimental results indicate that G-FLEX achieves over 99
In this paper, a new insight into the link between bit-oriented and word-oriented stream ciphers from a cryptanalytic perspective is given. That is, for a given bit-oriented stream cipher with a key size of n bits, it can be transformed to a word-oriented stream cipher with a key size of n-m bits, where a word consists of 2^m bits, the bitwise XOR and table lookup operations replace the bit XOR and AND operations respectively. Based on the new insight, a new cryptanalytic technique on bit-oriented stream ciphers is proposed. The new attack combines the guessing strategy with algebraic technique, and aims at slightly improving the exhaustive key search. To demonstrate the effectiveness of transformation attack, we apply it to the bit-oriented stream cipher ACORN v3.
Distributed key generation (DKG) is a crucial tool to achieve decentralized cryptographic functions such as distributed signatures and generating random beacon. Although numerous DKG protocols have been proposed, most of them only support the key generation of group elements rather than field elements, which is incompatible with the Chinese cryptography standard SM2 algorithms. In this paper, we first propose a distributed key generation protocol for SM2 algorithms. This protocol allows for the key generation of field element shares and eliminates the costly complaint rounds that greatly dominate communication cost in DKG. Further, as a somewhat direct application, we apply the keys generated from the DKG protocol for SM2 to design a threshold SM2 encryption scheme. Unlike previous threshold SM2 encryption schemes, our scheme does not require any trusted authority to share the key in advance and is proved secure under the malicious adversary model.
Mobile-app identification over encrypted network traffic is crucial in managing network, ensuring cybersecurity, and analyzing advertising. Combining machine learning classifiers and traffic features is the mainstream approach, which always assumes the training and test traffic is independent and identically distributed. However, in many real scenarios, the test traffic could dynamically change and drift out of the training distribution, resulting in obvious performance degradation. Existing methods recollect the drifted samples, and then minimize the differences between distributions or directly retrain the classifiers from scratch. However, collecting drifted traffic is labor-intensive and cannot cover all possible drifted scenarios, which restrict the deployment. In this paper, we propose a Flow causal Neural Network (FCNN) to improve the performance in identifying mobile apps in drifted scenarios under zero-recollecting. The FCNN concentrates on more robust associations between features and labels which persists invariant for drifted samples, and mitigates spurious correlations which is unstable via reweighting training traces. In addition, FCNN expands the seen feature space by randomizing the discriminative features to enhance the robustness against potential distribution drifts. In the extensive experiments on two public datasets, our FCNN achieves a remarkable improvement (4.68 ∼ 23.56 ↑ in F_1 ) with zero-recollecting and outperforms other comparisons.
