Quishing (QR code phishing) represents an emerging cybersecurity threat characterized by its dual-modal nature. This attack simultaneously exploits the visual modality of QR code images and the semantic modality of embedded URLs to launch covert attacks. Such a dual-modal attack strategy renders existing single-modality detection methods (relying solely on either images or URLs) and traditional fusion approaches based on expert-driven feature engineering fundamentally inadequate for comprehensive defense. To address this gap, this study proposes QRURL-GAAFNet (QR Code and Associated URL Features Global Attention-Based Adaptive Feature Fusion Network), which, to the best of our knowledge, represents the first end-to-end deep learning framework that holistically addresses the unique cross-modal correlation patterns inherent in Quishing attacks. Within this framework, dedicated deep learning networks independently extract visual features from QR codes and semantic features from URLs. Unlike conventional multimodal fusion approaches that treat modalities independently or use fixed fusion strategies, our proposed Global Attention-based Adaptive Feature Fusion (GAAF) mechanism dynamically learns sample-specific cross-modal dependencies by computing adaptive fusion weights from joint representations. By emphasizing discriminative information based on learned cross-modal correlations, this adaptive weighting strategy enables the model to focus on attack-relevant signatures that emerge from the interaction between QR code visual artifacts and URL semantic obfuscation patterns. Experimental results on public dataset demonstrate that QRURL-GAAFNet outperforms state-of-the-art methods in Quishing detection. Moreover, comprehensive experiments on the effectiveness of the data increment mechanism and robustness analysis, including evaluations against QR code image perturbations and URL text obfuscation attacks, validate its strong practical applicability in real-world security scenarios.
In recent years, phishing attacks have grown exponentially in scale, frequency, and sophistication, placing a significant burden on organizations and security personnel. Attackers leverage advanced obfuscation techniques to evade detection systems and ensure their emails reach users’ inboxes. Additionally, attackers distribute thousands of emails as part of large-scale coordinated campaigns to increase the probability of a successful attack. The growing use of social engineering and AI-generated phishing content has further heightened the complexity of these threats, making detection more challenging. As attackers continuously refine their tactics, mitigating these threats requires substantial time, expertise, and resources from IT/SOC teams, who must analyze, categorize, and respond to a high volume of evolving phishing attacks. In this study, we explore the integration of artificial intelligence (AI) into phishing mitigation, not as a replacement for human expertise but as a complement to it. We propose a novel hybrid pipeline designed to enhance the efficiency of IT personnel in responding to phishing threats. Our approach consists of three key components: (1) Feature Extraction, where fine-tuned language models analyze phishing emails to extract contextual features; (2) Campaign Detection, leveraging community detection algorithms to identify and cluster similar phishing emails; and (3) Campaign Profiling, generating comprehensive attack summaries to streamline response efforts. By clustering related emails and providing a comprehensive attack summary, our framework enables security teams to identify and neutralize threats faster, block incoming emails, and respond to users.
The sharing of image data in internet of vehicles (IoVs) plays an important role for traffic management in intelligent transportation systems. Although the emergence of cloud computing provides the storage of image data in the cloud server under internet of vehicles, a single cloud server may be vulnerable to some attacks, and there are computing costs and network delays caused by centralized communications. Therefore, in order to efficiently and securely share image data in internet of vehicles, we propose a attribute-based traceable image data sharing scheme under the setting of multi-fog-IoVs. First, the Shamir secret-sharing method is utilized to store image data distributed on the fog servers, where ciphertext-policy attribute-based encryption (CP-ABE) is used for encrypting image data. Second, the pixel values of image data are hashed to generate the label information for detecting the integrity of image data. Further, considering that dishonest vehicles (users) may maliciously release shared image data to unauthorized users, the zero-watermark technology (discrete cosine transform) is used to trace malicious users. Finally, we analyze the efficiency and security of our scheme, and related experiments show our scheme is efficient under multi-fog-IoVs.
Many Internet of Things (IoT) deployments rely on multi-hop routing protocols designed for low-power and resource-constrained devices. These protocols are susceptible to routing attacks that degrade network reliability, compromise data integrity, and disrupt system operation. Existing countermeasures typically address limited attack classes or rely on cryptographic mechanisms that impose prohibitive overhead on constrained hardware. This article presents a lightweight authentication framework to secure the IPv6 Routing Protocol for Low-Power and Lossy Networks (RPL) against combined Sybil and flooding attacks. The proposed approach constructs a global one-way hashchain and partitions it among provisioned nodes such that each index interval is associated with a specific node identity. This structure enables verifiable authentication of RPL control messages using only hash evaluations, eliminating the need for public-key operations during routine authentication. To maintain efficiency under adversarial load, the framework integrates probabilistic verification and per-node rate limiting to bound computational overhead during flooding attempts, while a Sybil pre-filter discards fabricated identities prior to full hashchain verification. For long-term deployments, credential renewal is protected using a Physical Unclonable Function (PUF)-gated mechanism that provides hardware-assisted identity verification before renewed credentials are distributed. The formal analysis demonstrates that the scheme provides message unforgeability, replay resistance, PUF-gated renewal security, Sybil resistance, and bounded verification cost under the stated cryptographic assumptions. A proof-of-concept implementation in Contiki-NG, evaluated on emulated Zolertia Z1 motes using Cooja, shows that the framework introduces 20 bytes of message overhead and less than 1% CPU utilization under benign conditions. Under the evaluated attack rates, the complete framework reduces radio duty cycle by 31–76% relative to unprotected RPL. Comparative analysis against symmetric MAC, ECC-based signing, ECC-voucher, and Merkle-hashchain approaches indicates that the proposed design provides a favorable balance of memory usage, latency, communication overhead, scalability, and attack resilience under the evaluated conditions.
With the rapid development of the Internet of Things (IoT) and artificial intelligence (AI) technologies, edge computing emerges as a crucial computing paradigm. By processing data near the source, edge computing enables faster and more efficient intelligent services. However, edge devices usually limit computational resources, and existing DNN inference latency profiling methods often rely on internal model details or large-scale latency measurements, making them costly and unsuitable for black-box deployment scenarios. This article proposes ProfEdge, a fast construction framework for deep neural network (DNN) latency profiling models based on Gaussian process regression and Bayesian optimization. ProfEdge adaptively samples real latency measurements under different batch-size states and tunes profiling-model hyperparameters to reduce construction cost while improving profiling accuracy. Specifically, ProfEdge builds an adaptive sampling module based on Gaussian process regression to locate high-error regions through coarse-grained sampling and dynamically refine the sampling process. It further designs a dynamic Bayesian optimization mechanism to improve the accuracy of the latency profiling model. Finally, ProfEdge constructs a cross-device performance mapping model to migrate an existing profiling model to a target device with lightweight stratified calibration, thereby avoiding full reconstruction of the target-device profiling model. Experiments on various edge devices and DNN models, including CNN-based and transformer-based workloads, show that ProfEdge reduces profiling errors by up to 80% and saves over 70% of profiling construction cost compared with existing methods. The cross-device migration results further demonstrate that ProfEdge can achieve competitive profiling accuracy on new devices with only a small number of target-device calibration samples.
