Wireless weak-link sensor networks (abbreviated as “weak networks”) are widely deployed in extreme environments such as underground, deep sea, and pipelines. Affected by physical characteristics including strong attenuation, high noise, and intermittent connectivity, their communication links persistently suffer from high packet loss, asymmetry, low bandwidth, and frequent disconnections. As a result, the security mechanisms of traditional wireless sensor networks struggle to remain effective in such environments. External attackers can exploit the vulnerabilities of weak network links to carry out identity forgery, man-in-the-middle attacks, and replay attacks, while compromised internal nodes may undermine network stability through impersonation, selective forwarding, and collaborative attacks. Therefore, constructing a security mechanism tailored to the characteristics of weak networks is a core issue in ensuring their usability.To address the above challenges, this paper proposed a security framework for weak network environments. For external access security, a disconnection-robust authenticated key agreement protocol (D-ADH) is designed. It significantly reduced interaction overhead by using fixed negotiation public keys and a single secure broadcast, and employed a lightweight request-retransmission mechanism to improve the success rate of key negotiation in high packet loss environments. For internal malicious node detection, a dynamic trust management mechanism based on Type-2 fuzzy logic (FDTM) was proposed. It integrated multiple sources of trust evidence, including communication success rate, data consistency, and traffic behavior, and introduced environment-aware and trend-based update methods, enabling trust inference to effectively distinguish anomalies caused by weak network noise from real malicious behaviors. The proposed security framework can simultaneously achieve stable external authentication capabilities and high-precision internal malicious node identification under extreme link conditions, providing a deployable, scalable, and highly robust security solution for weak networks.

As deepfake generation quality increasingly approaches that of real images, deepfake detection has become crucial for multimedia content security. However, most existing methods rely on statistical discrepancies between real and fake samples, rendering them potentially vulnerable under black-box conditions. This paper investigated the security of deepfake detection systems in a training-free, zero-query black-box setting and introduced a novel attack perspective: modeling adversarial attacks as targeted correction of statistical fingerprints in fake images, rather than exogenous noise perturbations. Based on this insight, this paper proposed SpectralFusion, a training-free black-box method that aligns frequency-domain distribution distributions. Leveraging the inherent real-fake paired prior present in deepfake generation, SepctralFusion identifies statistical discrepancies between fake images and their real references through frequency-domain analysis and applies controlled corrections only to anomalous frequency bands—without accessing model parameters, gradients, or additional training and queries. Specifically, we designed a difference-aware frequency band mask to accurately localize abnormal frequency components, and introduced an adaptive fusion strength mechanism to dynamically regulate correction intensity. Combined with a local overlapping block-based frequency processing strategy, our method enables fine-grained alignment and reconstruction of manipulated frequency features. Extensive experiments results show that SpectralFusion consistently deceives multiple deepfake detection models while preserving high visual fidelity, and generalizes well across diverse model architectures and manipulation types. Our findings reveal inherent vulnerabilities of deepfake detectors in the frequency-domain statistical space, offering a new perspective for evaluating the robustness of black-box detection systems in real-world scenarios.

The scale and complexity of various data management systems continue to escalate, with the data they generate exhibiting characteristics such as high dimensionality and non-linearity. This often leads to issues such as data redundancy, contamination by anomalies, and diminished quality, thereby posing a threat to system security. Traditional anomaly data detection methods struggle to address these challenges effectively, exhibiting limitations such as low accuracy and poor adaptability. This paper proposed an anomaly data detection approach based on quality dissimilarity. The method employed a particle swarm optimisation algorithm to identify key features and constructed a detection model by evaluating sample differences through data quality dissimilarity, thereby effectively distinguishing normal from anomalous data patterns. Experiments demonstrate that this method outperforms traditional statistical and machine learning approaches, particularly when handling high-dimensional, non-linear datasets. By characterising sample dissimilarity through data quality divergence, the method significantly enhances the accuracy and reliability of anomaly detection in complex environments.

This paper proposed a safe optimization algorithm for zero-sum game of nonlinear cyber-physical systems based on model-free reinforcement learning, specifically targeting active suspension system in vehicles subjected to denial-of-service attacks. The algorithm aimed to address safety control issues in scenarios with unknown system models and network packet loss. By introducing a Bernoulli random sequence to characterize the packet loss process caused by denial-of-service attacks, the attacked system was modeled as a stochastic nonlinear system. A discounted cost function incorporating control effort and disturbance penalty was defined, transforming the security control problem into a zero-sum game. A model-free value iteration algorithm based on Q-learning was designed, which constructed a Q-function involving state, control, and disturbance to avoid reliance on the system model. Furthermore, a neural network-based evaluation execution interference architecture was adopted to achieve function approximation. The evaluation network was used to approximate the Q-function, and the execution network and interference network were used to generate control strategies and disturbance strategies. Theoretical analysis demonstrates that the proposed algorithm ensures monotonic convergence and uniform boundedness of the value function sequence. Simulation results indicate that the method effectively maintains the stability and control performance of the suspension system even under denial-of-service attacks.

In the context of the continuous innovation and wide application of artificial intelligence technology, deep learning has achieved significant breakthroughs, but it often shows vulnerability when subjected to adversarial sample attacks. In the field of natural language processing, the discreteness and semantic constraints of text make text adversarial attacks more challenging. Among them, word-level attacks, as a typical combinatorial optimization task, existing methods face the dual bottlenecks of easily causing semantic deviation in the construction of the search space and often getting stuck in local optima of the optimization algorithm, making it difficult to efficiently explore high-quality perturbation strategies. To address these issues, this paper proposed a word-level adversarial sample generation method based on the improved slime mold algorithm. Firstly, by constructing a search space of replacement candidate words based on semantic origin knowledge, it effectively constrained the semantic consistency of the replacement words, avoiding the problem of naturalness degradation of the text caused by semantic deviation in traditional methods; then, the improved slime mold algorithm reconstructed the position update rule through discrete space logical operations, achieving a dynamic balance between global exploration and local development, avoiding the problem of the algorithm getting stuck in local optima; finally, by using the substitution strategy of similar characters in Chinese and Japanese dictionaries, the attackability of the adversarial samples was further enhanced. Experimental results show that this method achieves effective attacks on text classification models in the Chinese sentiment classification task, with an average decrease in classification accuracy of more than 30%, and significantly outperforms the comparison methods in semantic similarity.

For the data access control problem of edge intelligent controllers in industrial Internet of things scenarios under edge computing, this paper proposed a multi-authority policy-hidden attribute-based encryption scheme for edge intelligent controllers. This scheme achieved multi-authority key generation through the collaboration of a central authorization authority, a key generation center, and the edge intelligent controllers. It also employed a one-way anonymous key agreement protocol to achieve full policy hiding. To improve encryption efficiency, the scheme integrated online/offline techniques and employed outsourced decryption technology, delegating most of the ciphertext computation tasks to edge servers, thereby reducing the computational overhead for users. In addition, the scheme also featured user tracking functionality, allowing the tracking and revocation of malicious user based on the decryption key. Analysis results show that the proposed scheme exhibits high performance in user key generation and file encryption/decryption, and its security is validated under the q-DBDHE assumption.

With the widespread adoption of encrypted communications, traffic classification faces new challenges. Traditional methods perform poorly when dealing with encrypted traffic, as existing approaches either rely on manual feature extraction or fail to fully capture the interaction patterns between packets. To address this issue, this paper proposed a Multi-feature Fusion based Encrypted Traffic classification method (MFF-ETC). In the preprocessing stage, the method combined packet-level images generated from packet payloads into session images, effectively mitigating information confusion while preserving the interaction patterns among packets. In the classification stage, the session images were processed by three modules: the Packet Vision Transformer (PVT), the Temporal Traffic Convolutional Network (T-TCN), and the Traffic Gated Bottleneck Convolution (T-GBConv) module, which extracted global features, full-scale temporal features, and spatial features, respectively. Subsequently, a dynamic weighting mechanism fused these three types of features, adjusting their weights adaptively according to the traffic type to achieve more accurate classification. Experimental results demonstrate that MFF-ETC achieves F1-score of 98.81%, 98.93%, and 98.05% on the ISCX-VPN-Service, ISCX-VPN-App, and CSTNET-TLS1.3 datasets, respectively, validating the method’s high classification accuracy and generalization capability.

A multi-party collaborative blind signature scheme based on State Secrets SM9 was proposed to address the issue of collaborative signatures in multi-user scenarios that cannot be solved by existing State Secrets SM9 blind signature schemes. The plan included six steps: system initialization, key extraction, message blinding, signature, unblinding, and verification. Firstly, the signature private key was divided and distributed to all signatories by the key generation center. After the message owner blinded the message, all signatories completed the signature scheme to generate a valid SM9 blind signature, which was then unblinded by the message owner. Finally, the SM9 verification algorithm was used to validate the signature. The security of the signature scheme has been proven under a universal architecture, with signature lengths close to SM9 signature schemes and computational costs lower than existing collaborative signature schemes. This scheme ensures that all parties involved in the signature cannot obtain message privacy and will not disclose their private keys during interaction, meeting the requirements of unforgeability and blinding. It solves the practical needs of multi-user blind signature scenarios and achieves the security of multi-party collaborative signatures and collaborative processes.

Factual hallucination refers to the phenomenon where large language models generate text that is contextually inconsistent or contradictory to established facts during the text generation process. This issue not only undermines the reliability of the generated text but may also lead to the dissemination of misinformation, thereby negatively impacting user decision-making. To address this challenge, this paper proposed a novel hallucination mitigation method based on a Multi-View Knowledge Graph (MVKG) and Direct Preference Optimization (DPO). The method introduced MVKG and employed the large language model itself to update the knowledge graph. Simultaneously, it adopted a multi-view keyword joint retrieval mechanism, combined with multi-view direct preference optimization, to enhance the model’s ability to extract multi-view entities from queries, thereby improving the overall relevance of retrieval. Experimental results demonstrate that, compared to existing Retrieval-Augmented Generation (RAG) methods, the proposed method significantly improves question-answering accuracy on both the Qwen series of models and closed-source models, effectively mitigating the factual hallucination problem in large language models.

This paper presented a confidential federated learning platform (CFLP) based on trusted execution environments (TEEs) for typical vision tasks, aiming to evaluate the security-efficiency-accuracy trade-off of different privacy-enhancing technologies in federated learning. The platform utilized intel trust domain extensions (TDX) and software guard extensions (SGX) as its core architecture, while incorporating homomorphic encryption (HE) and secure multi-party computation (MPC) as performance comparison benchmarks. Systematic comparative experiments were conducted using this platform in high-dimensional vision task scenarios involving the CIFAR-10 dataset and the ResNet-18 model. The results indicate that, while maintaining baseline accuracy, the TDX-based TEE scheme provided virtual-machine-level hardware protection with only an approximately 1.3% increase in end-to-end latency, outperforming SGX, HE, and MPC in comprehensive performance. Although HE offers formally verifiable security, it increased the single-round training latency and communication overhead to approximately 9 times and 21 times that of the baseline, respectively, resulting in significant computational overhead. MPC exhibited limitations in the trade-off between time and communication costs. This study clarifies the applicable boundaries of various technical solutions, demonstrating that for secure aggregation scenarios involving high-dimensional models, TDX is a favorable option for balancing security requirements and performance overhead.

Efficient management and retrieval of network information represent a critical research topic in large-scale cyberspace mapping. Facing the explosive growth of internet information, rapid web content parsing and accurate topic identification become significant challenges. Therefore, this paper proposed an efficient web page topic classification method based on fine-tuning TinyBERT and an improved TextCNN. The study first employed TinyBERT, a lightweight pre-trained knowledge distillation model with advanced attention mechanisms and strong semantic understanding capabilities, as an encoding module. Fine-tuning this model captured contextual semantic features of web pages efficiently. Then, the researchers introduced an improved TextCNN module to perform multi-scale semantic extraction on encoded features through groups of heterogeneous convolutional operations. Finally, a high-dimensional feature abstraction module was designed to fuse contextual semantics with multi-scale features. This process generated deeper feature representations and enhanced classification accuracy significantly. Experimental results on the WebKB and THUCNews datasets demonstrate that the model outperforms existing state-of-the-art web topic classification methods. The proposed approach achieves superior performance in accuracy, F1-score, and inference efficiency.

With the rapid development of new technologies, the complexity of security protocols has been increasing, posing unprecedented challenges for formal analysis. In recent years, automated verification technology has made significant progress in the field of security protocol security testing, becoming a hot research topic. This paper proposed a PCL-based HZ logic and its automated verification method, and demonstrated the effectiveness of the proposed method in verifying the confidentiality and authentication of communication protocols through the application case of the provably secure and lightweight authentication (PSLA) protocol. By constructing an attacker model, this paper simulates eavesdropping attacks, man-in-the-middle attacks, and replay attacks, and automatically verifies the security of PSLA under different attack scenarios. This provides an efficient and automated solution for the security analysis of secure authentication protocols, enhancing the security and practicality of protocol design.