By distributing transactions across different shards for parallel processing, sharding improves the scalability of blockchain systems. When transactions involve multiple shards, coordinated execution is required, making the efficiency of cross-shard transactions (CTXs) a critical determinant of overall system performance. However, existing CTX protocols lack pre-execution state verifiability in the mempool, forcing target shards to defer transaction processing until source shard confirmation. Moreover, the tightly coupled source and target execution pipeline serializes CTXs processing across shards, significantly restricting parallel execution of CTXs. To address these issues, this paper proposes ParaShard, a high throughput and low latency CTX protocol based on a fast path parallel execution. ParaShard decouples transaction acceptance from on-chain commitment by enabling source shards to generate off-chain threshold signature lock proofs, which can be directly verified by target shards. As a result, target shards can admit CTXs into their local transaction pool without waiting for source shard consensus, enabling early processing and parallel execution across shards. Moreover, we implement a prototype of ParaShard and evaluate it using historical Ethereum transactions. Experimental results demonstrate that, compared with state-of-the-art protocols such as X-Shard and Monoxide, ParaShard achieves 30.5
Mobile access within public finance and enterprise environments often requires lightweight anonymous authentication, allowing users to prove authorization without disclosing their identities. Group Time-based One-Time Passwords (GTOTP) has recently been proposed as a lightweight primitive meeting this need with post-quantum security. To address dynamic group membership, Cao et al. introduced DGTOne, the first dynamic GTOTP construction. It employs chameleon hashes to precompute a fixed set of Merkle-tree leaves (mount points), into which conventional TOTP verification points (VPs) contributed by group members are adaptively inserted. However, DGTOne partitions mount points by time epochs, so they can expire and become unusable, causing capacity waste due to unpredictable join times. Moreover, its outsourced proof generation requires verifiers to be online each epoch to fetch refreshed credentials from Registration Authority (RA), defeating offline verification needed in mobile access. We address these limitations with two new schemes. First, we propose , a no-wastage DGTOTP design that constructs Merkle trees of members’ verification points (VP-trees) on demand, eliminating expired mount points at the cost of added handling latency. To mitigate this latency, we introduce , which instantiates multiple small one-time signature (OTS) trees whose leaves (OTS public keys) serve as mount points. New members’ VPs are signed immediately using unused leaves, achieving low wastage. We formally prove that the wastage rate of is, with overwhelming probability, lower than that of DGTOne. By modeling the registration process and optimizing OTS-tree size, for deployments with up to 500 members (209 initially, 20 added monthly), reduces mount point wastage rate by 10.2
In this work, we propose a novel image steganography framework based on adaptive embedding of a secret image into a cover image using a Convolutional Neural Network (CNN) and Generative Adversarial Network (GAN). The system dynamically selects optimal embedding zones and enhances the visual and structural fidelity of the reconstructed secret image Unlike conventional approaches that rely on fixed embedding regions or uniform coefficient allocation, the proposed approach adopts a content-aware, block-based embedding strategy that jointly optimizes imperceptibility, robustness, and payload capacity. For each pair of secret and cover blocks, transformed and quantized frequency coefficients are analyzed by a trained CNN to identify the most suitable insertion area based on texture complexity and spectral characteristics. To ensure confidentiality and strengthen resistance to steganalytic attacks, the coefficients of the secret blocks are encrypted using a lightweight and efficient algorithm prior to embedding, guaranteeing that the hidden information remains unintelligible even if successfully extracted. Additionally, the GAN-based refinement module is used to enhance the visual quality of the restored secret image. Experimental evaluations demonstrate that the proposed model achieves high stego-image quality, strong resilience to compression and noise attacks, and accurate secret-image reconstruction.
As communication infrastructures migrate towards post-quantum cryptography, Mersenne number-based AJPS cryptosystems have emerged as promising lightweight candidates for resource-constrained environments, including IoT networks. These schemes offer high arithmetic efficiency, yet their security relies heavily on the assumed hardness of inherent mathematical problems. In this paper, we evaluate the resilience of these primitives against advanced cryptanalytic threats. We propose novel attacks that leverage continued fractions and lattice-based strategy for modular polynomial equations. Crucially, we demonstrate that the computational complexity of our proposed attacks is polynomial in the bit-length of the modulus and is independent of the keys’ Hamming weight. We significantly relax the size constraints on the keys, proving that the weak key space is much larger than estimated. We validate these findings through extensive numerical experiments, confirming that secret keys can be recovered in seconds. Our results indicate that current parameter selections for Mersenne number-based AJPS schemes are insufficient for long-term data protection, necessitating a rigorous re-evaluation of security margins.
We present Sinba, an optimized realization of Vectorized Batch Private Information Retrieval (VB-PIR) built on RLWE-based homomorphic encryption with SIMD batching. While prior VB-PIR systems significantly reduce communication by packing queries and merging responses across buckets, their server-side online latency is often dominated by rotation-heavy packed inner products executed multiple times (e.g., slot selection and aggregation). Sinba targets this bottleneck through three complementary techniques. First, it encodes database records into a two-dimensional slot layout, mapping multi-slot payloads onto multiple plaintext rows with regular structure. Second, it rewrites the dominant rotate-multiply-accumulate computation as a 2D baby-step/giant-step (BSGS) schedule, enabling reuse of baby-step rotations across rows and supporting an unbalanced parameter choice tuned to the number of rows, thereby reducing rotations from (N) to approximately (√(N)) . Third, Sinba shifts rotation alignment into offline preprocessing, storing the encoded database in NTT form to accelerate ciphertext-plaintext products in the online phase. We evaluate across batch size, database size, and payload length. On a database of 2^20 items with 256-byte payloads, Sinba reduces server response latency by 8.5-13.3× compared to the VB-PIR baseline of Mughees and Ren (IEEE S P 2023), while preserving the communication advantages of vectorization.
To ensure cryptographic agility and that security protocols still remain effective, in retaining information security even against increasingly complex threat models, transitioning conventional cryptosystems to new secure quantum proof ones requires more than a mere substitution of the internal algorithms: It mandates an elegant solution maintaining the functional richness enhanced with quantum-secure constructions that allow the smooth migration to long-lived trusted digital infrastructures without requiring substantial system overhauls. Despite the maturity of lattice structures (based on the “Learning with Errors” (LWE) problem) which serves as the foundation for a plethora of cryptographic applications, their practicality still remains nascent due to their deliberate design constraints on operating with linear structures and inherent quadratic overhead. A main open question arises: Can the LWE problem and its applications be generalized with algebraic structures that allow for the design of new efficient cryptographic schemes supporting all core key management operations, including key encapsulation and distribution? We resolve this question in the affirmative by presenting a new construction that augments LWE with invertible secret keys and supports secret-key composable operations (multiplication) via polynomial matrices, retaining standard security assumptions. To demonstrate the practicality of our construction, we develop a two-party decryption scheme that achieves distributed functionality without the overhead of traditional “trapdoors” and gadget-based primitives. Furthermore, we prototype and integrate our approach with ML-KEM (Kyber), observing comparable performance. This serves as evidence on both the performance and crypto-agility of the presented construction, further advancing lattice-based cryptography with field-like operational semantics that better fulfill overarching (application-driven) requirements.
Issuer-Hiding Anonymous Credential (IHAC) is a cryptographic primitive, which enables users to authenticate themselves with fine-grained control over their credentials (certified attributes), without revealing the credential issuers. A state-of-the-art IHAC scheme, designed for Self-Sovereign Identity systems, was recently published by Sanders and Traoré at PET’24. However, this scheme does not support the revocation of credentials. It is well-known that revocation is essential for many secure applications. In this work, we introduce a revocable IHAC scheme. Our contributions are three-fold. First, we use the classical pairing-based accumulator with Sanders and Traoré’s scheme to provide user revocation without compromising its unique features like compactness and issuer-hiding. Second, we prove the security of our construction under the well-studied q-Strong Diffie-Hellman (q-SDH) assumption, rather than the Generic Group Model (GGM) originally used by Sanders and Traoré. Third, our construction ensures optimal efficiency, as the accumulator-based revocation incurs only a small extra overhead to PET’24. Our implementation shows that when using 100 certified attributes and 20 issuers in an anonymous credential system, the and protocol takes 3.15 s in total, only incurring a 2.5 % increase compared to Sanders and Traoré’s scheme.
Cross-chain bridges have become critical infrastructure for today’s blockchain ecosystem, routinely carrying assets worth hundreds of millions of dollars across heterogeneous chains. Yet many of the most damaging bridge failures do not stem from broken cryptography, but from mismatches between what a proof or message is assumed to guarantee and what the destination contract actually verifies. These authorization-time assumptions are usually implicit, scattered across code, configuration, and informal documentation, and are therefore hard to audit, compare, or reason about systematically. We argue that secure cross-chain communication requires an explicit, long-lived, machine-readable semantic interface at the authorization boundary. We present an ontology-driven interface that models messages, evidence, routes, headers, runtime policy, and context as first-class concepts, organized into reusable semantic anchors for domain binding, finality, ordering, uniqueness, and context compatibility. On top of this ontology, we build a SHACL-based diagnostic layer that classifies, for each evidence package, whether each anchor is evidenced, contradicted, or not established. Using (i) a controlled mutation regression suite that targets each semantic anchor family and (ii) incident-aligned interface snapshots for three widely documented bridge incidents (Nomad, Wormhole, and Poly Network), we show that the resulting “assumption graphs” can localize semantic failures such as misconfigured trust roots, out-of-policy context, and route/domain misbinding—often without inspecting contract code. From these controlled and incident-aligned reconstructions, we derive a reusable interface template: a minimal set of fields and invariants that make authorization-time assumptions explicit, auditable, and comparable across bridge designs. Our results suggest that many catastrophic bridge failures could have been diagnosed earlier, had their authorization semantics been made visible at the interface level.
Scalable off-chain payments are crucial for supporting high-frequency transactions in payment channel networks (PCNs). In the Lightning Network, this is typically achieved through multi-path payments that split a payment across multiple paths to improve transaction success rates. However, existing designs impose significant communication overhead on the payer by requiring individualized encrypted information for each intermediate user. Moreover, onion-routing-based mechanisms, which are widely adopted for hop-by-hop forwarding in PCNs, are inherently designed for single-path communication, making them unsuitable for efficient multi-path data delivery and end-to-end verifiability. To address these limitations, we present DiVeR, a cryptographic framework for secure data dissemination in multi-hop, multi-path payment networks. To enable differential data delivery, DiVeR leverages a key-tree–based derivation mechanism to assign distinct encryption keys to nodes on different payment paths. At the same time, DiVeR employs chained ring signatures to provide strong end-to-end data verifiability between the payer and all recipients. In addition, we design a multi-layer encrypted packet that extends the self-routing property of onion encryption to multi-path settings. We implement DiVeR and evaluate it against state-of-the-art schemes. Experimental results show that DiVeR achieves on average 2.25 × higher throughput than AMHL and 1.38 × higher throughput than Rollercoaster and PolySphinx.
Selfish mining enables attackers to increase their revenue by hiding blocks and strategically releasing them, thereby wasting the computational resources of honest miners. While many countermeasures rely on protocol-level modifications that are difficult for individual miners to adopt, insightful mining was proposed as an individual-level defensive strategy and was shown to dominate selfish mining when its mining power exceeds one third. In this paper, we examine the effectiveness of insightful mining against equal-fork stubborn mining (F-stubborn mining), a stronger attack that extends selfish mining by adopting a more aggressive block-reveal policy. We show that, under equal mining power, insightful mining achieves a slightly higher relative revenue than F-stubborn mining. However, this advantage is small and quickly diminishes once mining power becomes asymmetric. Especially, when the F-stubborn pool holds a power advantage, its revenue increases rapidly and approaches a majority share as its mining power nears one half. These results indicate that the defensive capability of insightful mining against stubborn mining is inherently limited.
Two-factor authentication (2FA) is now a baseline for web and app accounts, but deployments remain vulnerable to device theft, social engineering, and adversary-in-the-middle infrastructures that trick users into relaying one-time passwords (OTPs) or approving login prompts they did not initiate. We explore an environmental third factor that is explicitly tied to physical context. Our scheme instruments enclosed spaces, such as homes or offices, with near-ultrasonic beacons that broadcast digitally signed messages. A local server, reachable only over local Wi-Fi, maintains a queue of recent beacons. After a conventional 2FA check succeeds, the client verifies that at least n of M expected beacon signatures are present in this queue before finalizing authentication, adding a location-bound check that runs transparently in the background whenever users log in from covered environments. We design and implement this scheme on commodity hardware and evaluate a 2-of-4 deployment in a lab setting that approximates an office, analyzing beacon verification patterns, bit error rates, and recovery time after transient drops in the number of valid beacons. We further report on a 40-participant user study with 160 authentication attempts that compares baseline TOTP-based 2FA to our augmented scheme, combining task-level measures with System Usability Scale (SUS) scores and qualitative feedback to assess how an always-on environmental factor fits into usable authentication.
Inner product functional encryption (IPFE) enables secure computation on private data for statistical analysis but inherently leaks information about encrypted data, necessitating fine-grained access control. Existing solutions use attribute based encryption to restrict who can recover the function output, enforcing only unilateral receiver-side control. However, real-world data sharing often requires more complex and flexible control: not only who may decrypt, but also from whom a receiver may accept data. We address this by proposing a matchmaking IPFE scheme with bilateral authorization and attribute hiding. Inspired by matchmaking encryption, our construction introduces a dual inner product predicate that simultaneously matches sender and receiver attributes. This predicate is embedded in the IPFE evaluation, ensuring the function value is revealed only when both parties satisfy the matching condition. The scheme achieves semi-adaptive simulation-based (SAD-SIM) security under standard assumptions. Experiments show modest overhead in key generation, encryption, and decryption, which remains low even as predicate and function vector dimensions grow, demonstrating practicality for real-world privacy-preserving analytics.
This research presents a novel framework, SHAP-RM, for assessing the reliability and trustworthiness of machine learning (ML) models in cyber security applications using explainable artificial intelligence (XAI). By applying SHAP (SHapley Additive exPlanations) to XGBoost regression and classification tasks trained on the live PV generation dataset and the UNSW-NB15 dataset, we examine the interpretability and robustness of model decisions. Integrating XAI enhances transparency, offering actionable insights for refining and securing ML-driven defense systems.
Hierarchical federated learning has emerged as a practical solution for privacy-preserving machine learning in healthcare, aligning with the multi-tier structure of Internet of Medical Things deployments. However, hierarchical federated learning remains vulnerable to inference-time adversarial attacks, which undermine model integrity at deployment, and harvest-now-decrypt-later attacks, which threaten the long-term confidentiality of encrypted model updates. To address these challenges, this paper proposes a secure and adversarially robust hierarchical federated learning architecture, Q-RobHFL. Q-RobHFL combines client-side adversarial training with quantum-secure edge–cloud communication. Model robustness is enhanced through adversarial training at the client level, while long-term confidentiality is ensured by integrating Quantum Key Distribution for edge–cloud key exchange. Combining hierarchical federated learning, adversarial training, and quantum key distribution is crucial, as it protects the architecture from threats at different layers. Experiments on medical imaging data show a 45
Fog computing extends cloud capabilities to the network edge, enabling low-latency support for smart-city and IoT applications. However, the heterogeneity and limited security capacity of fog and edge nodes make security-aware resource allocation a critical challenge. This paper proposes a Markov Decision Process (MDP)-based framework for allocating application modules across cloud–fog–edge infrastructures while considering security constraints. The allocation problem is modeled as a static decision process in which each module is assigned to a device using a composite reward function capturing security compliance, probability-weighted breach risk, and resource utilization constraints. The MDP is solved offline using value iteration to derive a reward-optimal static allocation policy. The approach is evaluated on a smart-city fog scenario comprising 25 application modules and 11 heterogeneous devices and is compared against a genetic algorithm (GA) baseline using an identical reward structure. Experimental results show that the MDP achieves 68
Android malware poses significant threats to mobile users by compromising privacy and security. Identifying malicious code within an application is essential for effective malware analysis and the development of countermeasures. This capability is particularly important for AI-based malware detection methods, as it enhances the interpretability and transparency of the models. Convolutional Neural Networks (CNNs) have shown great potential for malware detection due to their ability to capture spatial patterns in image-like representations of code. By converting code into visual formats, CNN-based approaches facilitate automated feature extraction and reduce dependence on manual analysis. While prior studies have focused on AI-based malware detection and generating heatmaps from malware images, our approach goes a step further by translating these visual explanations back to the corresponding code segments for malicious code localization. We propose a CNN-based framework that ensembles multiple saliency techniques (Grad-CAM++, Score-CAM, SmoothGrad, Vanilla Saliency, Integrated Gradients) to highlight critical regions in malware images and translate them into their corresponding code segments. The resulting heatmaps emphasize key parts of the code that contribute to malware classification. To enable pixel-level evaluation, we construct ground-truth pixel masks by mapping class-level malicious annotations to DEX byte ranges and then to image coordinates. Beyond localization accuracy, we aim to improve analyst usability by producing a ranked list of suspicious Smali classes and supporting artifacts (e.g., configuration/constants and resource references) that can be directly inspected. Experimental evaluation on real-world malware, using metrics such as Area Under Curve (AUC) and Normalized Scanpath Saliency (NSS), demonstrates that our approach accurately and effectively identifies malicious code regions.
Federated graph analytics has emerged as a vital paradigm for extracting insights from a large graph distributed across different data holders, where each data holder owns only a local subgraph. This paradigm is particularly vital in the financial sector, where transaction data of different institutions collectively form a global graph. Detecting structural patterns in a large federated transaction graph is highly valuable, greatly facilitating cross-institution anti-money laundering to combat illicit financial activities. Among others, the parallel chain pattern has been proven highly effective in uncovering sophisticated money-laundering behaviors. Detecting such a pattern in the federated setting is thus crucial. However, this is quite challenging since each data holder could be reluctant to directly share its local subgraph due to stringent privacy regulations and severe privacy concerns. This paper introduces OblivPCD, the first system framework supporting oblivious parallel chain detection in a federated graph setting. At the core of OblivPCD is a delicate synergy of insights on graph modeling and cryptographic computation. We implement OblivPCD and evaluate it on graph datasets containing tens of millions of vertices and edges. Experimental results show that OblivPCD greatly outperforms the state-of-the-art prior work in oblivious path extension. In particular, OblivPCD achieves up to a 33.5
Vehicular Crowdsensing (VCS) leverages built-in sensors of intelligent vehicles to collect data, showing great potential in fields such as smart cities and intelligent transportation. However, key challenges exist in practical applications: the reliability of data from individual vehicles is insufficient; traditional truth discovery methods inadequately consider the differences in vehicle credibility; and the limited platform reward budget struggles to effectively incentivize vehicle participation, especially in areas with low coverage. To address these issues, this paper proposes a truth discovery and incentive method oriented towards data coverage and data quality for crowdsensing. The method first designs an incentive mechanism by incorporating the Goal Gradient Effect from behavioral economics. It stimulates participation willingness by quantifying the gap between vehicles and their goals when selecting different tasks in remote areas. This mechanism encourages vehicles to perform tasks in low-coverage regions and grants them faster reward unfreezing speeds, thereby significantly improving the geographical uniformity of data collection. Additionally, vehicle trust values are evaluated based on known high-trust users, and the weighted aggregation process is optimized accordingly to enhance truth accuracy. Experiments demonstrate that the proposed scheme effectively optimizes data coverage while ensuring data quality.
Global carbon-neutrality policies are driving rapid growth in distributed renewables and peer-to-peer energy trading. However, existing blockchain-based trading systems still face three key challenges: negotiation privacy leakage, insufficient support for distribution-network constraints, and the tension between payment anonymity and regulatory accountability. In this paper, we present P ^2 NC-DET, a decentralized energy-trading framework that jointly addresses privacy-preserving negotiation, network-aware trade validation, and auditable anonymous payment. Specifically, we design an ECC-based private negotiation protocol to protect bids and participant identities during off-chain matching, while enabling the distribution network to validate whether the negotiated trade remains within pre-defined operational limits. We further extend Zether with auditor-recoverable account bindings so that payment amounts and counterparties remain hidden from the public, yet suspicious transactions can still be traced with the assistance of authorized entities. Security analysis shows that P ^2 NC-DET achieves identity privacy, transaction confidentiality, auditability, and resistance against common internal and external attacks under our stated trust assumptions. We implement a prototype on a private Geth network with Raspberry Pi nodes. Experimental results show that P ^2 NC-DET improves trading throughput and reduces negotiation latency compared with representative baselines, while incurring acceptable on-chain and cryptographic overhead.
With the rapid advancement of Internet technology, secure key distribution serves as a cornerstone of network communication. Traditional solutions often suffer from issues such as distinguishable traffic patterns, high computational complexity, or excessive cost. To address these challenges, this paper proposes a scheme for imperceptible key generation over the wide-area Internet—a transparent process to third parties. The scheme leverages round-trip time (RTT) as a side-channel entropy source to extract inherent Internet randomness, coupled with clock synchronization, implicitly establishing reciprocal signals during normal data interactions for sustained, long-range, and low-overhead key generation. To further enhance security and signal randomness, the scheme incorporates a multi-path parallel measurement mechanism that employs controlled active packet loss and random delay injection. On this basis, the key generation is finalized through equal-probability quantization, polar code-based forward error correction, and privacy amplification. Experimental results demonstrate that in both static and mobile scenarios, the initial key disagreement rate ranges between 10
