Data, the “new oil” of the digital age, must be shared with absolute security to unlock its significant value. Traditional data security sharing models suffer from centralization and trust challenges, and can prone to “single point of failure”. Blockchain has the characteristics of decentralization, transparency and immutability, and the integration of blockchain technology and data sharing can enhance its security and credibility. This paper provides an overview of blockchain-based data security sharing research. Initially, it presented a data security sharing model based on blockchain, and expounded the concepts and definitions of three key technologies: data storage, access control, data transmission and privacy protection. Subsequently, it offered a detailed analysis and comparison of the basic concepts, advantages and disadvantages, and applicable scenarios of research related to data storage, access control, consensus mechanism, secure transmission, and cross chain technology. Finally, we discussed key issues regarding blockchain-based data security sharing, such as the challenges of dynamically adjusting permissions, conflicts between data privacy and transparency, and offered some feasible solutions and future directions.

With the rapid development of Large Language Models (LLMs), their application potential in the public security domain has become increasingly prominent. However, issues such as insufficient capability transparency, over-alignment leading to unavailability, hallucination generation, and security threats hinder their ability to meet the high sensitivity, high risk, and high precision requirements of public security scenarios. This paper systematically reviews trustworthiness issues of LLMs in the public security context: it examines their current applications in tasks such as risk warning, security incident response internal management, and public services; defines trustworthiness and categorizes risks into internal vulnerabilities, external threats, and concomitant issues; and, based on the characteristics of the general domain, private network domain, and internet domain, proposes five trustworthiness dimensions—task suitability, factual accuracy, safe completion, adversarial robustness, and accountability. Following this structure, the paper surveys corresponding enhancement strategies and challenges, with the aim of promoting reliable, secure, and controllable applications of LLMs in the public security sector.

To evaluate the defense capability of capsule network against membership inference attacks, this study implemented membership inference attacks on the FashionMNIST and CIFAR-10 datasets and selected LeNet, VGG16, and ResNet18 as shadow models. Additionally, this study tested the impact of the number of shadow models on the attack effectiveness, explored the relationship between overfitting and membership inference attacks, and tested the defensive effect of differential privacy against membership inference attacks. The experimental results show that the attack success rate of membership inference attacks can reach up to 94.8%, and there is no significant advantage in the attack success rate when the number of shadow models is between 1 and 5. Furthermore, the study found that the effectiveness of membership inference attacks increased with the increase in overfitting, and the application of differential privacy technology can effectively enhance the defensive capability of the capsule network, but the training time of the capsule network will increase by more than 133%. These findings indicate that common strategies and defensive measures against membership inference attacks are applicable to capsule network, highlighting the importance of prioritizing security issues in the design and application of capsule network.

Federated learning mitigates the risks of centralized data storage through distributed training, yet remains vulnerable to malicious clients exploiting GAN attacks to steal private data. Traditional defense methods such as differential privacy and encryption mechanisms suffer from challenges in balancing model performance and privacy effectiveness or incur high computational costs. To address the threat of GAN attacks in federated learning for image recognition tasks, this paper proposes a privacy enhancement method based on Rényi differential privacy (RDP) to improve data privacy. The serial composition mechanism of Rényi differential privacy enables the privacy budget growth rate in multi-round iterations to transition from the linear scaling of traditional differential privacy to sublinear scaling, effectively reducing the amount of noise added. Thus, the method leverages the tight noise composition properties of RDP by incorporating gradient clipping based on weight equilibrium and optimized Gaussian noise injection into client-side gradient updates. This approach enables differential privacy-preserving computations, effectively reducing privacy leakage risks while balancing model utility. Experiments show that the method realizes local data privacy protection and enhances the privacy protection ability of the model under the premise that the degree of impact on the model’s global accuracy remains acceptable, so as to effectively resist GAN attacks and ensure the privacy of image data.

Aiming at the limitation of static security management mode in dealing with dynamic security management scenarios, considering the influence of offensive and defensive confrontation behavior on strategy selection, this paper put forward a security defense strategy selection method of information system based on dynamic security management mode. Combining belief theory, a belief random game model was constructed to effectively simulate the belief state and the attack and defense process of information systems in the face of different security threats. By analyzing the game relationship between them, the security state of the system was evaluated, and the defense costs and benefits of managers in the attack and defense state were calculated, as well as the impact on the success rate of attacks, so as to the optimal defense strategy. Taking the real classified information system as the research object, this paper demonstrated the effectiveness of the experiment from three aspects: attack success rate, defense cost and defense benefit, which provides scientific basis and improvement suggestions for the security management of information system.

This article aimed to explore the optimal time strategy for address hopping systems in moving target defense, and proposed a dynamic defense decision model based on game theory to address the asymmetry problem in network attack and defense. The research focused on three address hopping strategies: periodic, event driven, and hybrid driven, quantifying their defense benefits and comparing the effectiveness of strategy selection. By constructing a single-stage complete information zero sum game model, the defense success probability, jump frequency, and benefit function of different strategies were derived. The dimensionless method was used to experimentally analyze key parameters such as jump cost ratio, defense failure loss ratio, attack frequency. The results indicate that the Mixed strategy has the highest defense benefits in most scenarios, but the Fixed strategy has a greater advantage in high jump cost ratios or high defense failure loss ratios. This study provides a theoretical basis for the optimization selection of moving target defense time strategy, revealing the impact mechanism of the coupling effect of attack and defense parameters on strategy decision-making, thereby enhancing the adaptive ability of dynamic defense systems.

Low-Rate Denial-of-Service (LDoS) attacks exploit vulnerabilities in network protocols’ adaptive mechanisms, causing these mechanisms to fail in a legitimate manner, significantly reducing bandwidth utilization and quality of service. Therefore, the high concealment and destructive nature of LDoS attacks make them an important research topic in the field of network security.Aiming at the concealment of complex low-rate denial-of-service (LDoS) attacks across multiple network layers and the limitations of traditional detection methods in sampled traffic scenarios, this paper proposes an LDoS attack detection method based on HLD-Sketch (Hybrid-LDoS-Detect-Sketch). The study covers the detection of transport-layer LDoS attacks, application-layer LDoS attacks, and hybrid multi-layer attack under sampling conditions. First, an improved CM-Sketch structure is introduced to dynamically estimate flow lengths and adaptively adjust sampling probabilities, prioritizing fine-grained sampling for short flows to reduce interference from long-flow background noise during attack feature extraction. Second, leveraging the lightweight nature of CM-Sketch, multidimensional temporal statistical features, such as flow rate, the number of upstream and downstream packets, and port dispersion, are efficiently extracted from the sampled traffic Finally, a machine learning classifier is employed to hierarchically detect transport-layer, application-layer, and hybrid attacks. Experimental results demonstrate that the proposed method achieves a detection accuracy of 99.94% with a 3% sampling rate within 6 seconds, even in hybrid attack scenarios. This approach provides a lightweight solution for real-time detection of multi-dimensional LDoS attacks in high-speed network environments, particularly suited for resource-constrained scenarios with large-scale traffic.

To address the issue that traditional network defense decision-making methods are unable to effectively cope with complex dynamic network environments and diverse network attacks, this paper proposed a network defense autonomous decision-making method based on hierarchical reinforcement learning, combined with a high-fidelity network attack and defense simulation environment. A Markov network attack and defense game model based on incomplete information was constructed to analyze the dynamic interaction process of the attacker and defender and to formally represent the optimal defense strategy. The complex defense decision-making task caused by the unknown type of attacker was decomposed through the collaborative work of the top-level control agent and the bottom-level execution agent. Simulation experiment results under different attack and defense scenarios show that this method can make flexible and efficient decision responses to two types of penetration attack patterns, maintain resilient defense, and generate interpretable action distributions. Comparative analysis with existing related work further confirms the superiority of the proposed method in defense effectiveness.

The dynamic node participation and exit process can effectively enhance the flexibility in asynchronous federated learning (FL) environment. However, in scenarios involving data privacy and security, ensuring the legitimacy and secure exit of participating nodes is crucial. This paper proposed a lightweight dynamic node participation scheme for federated learning nodes supporting attribute update. Firstly, by introducing attribute-based encryption and revocation mechanisms, this paper designed a secure and flexible participation mechanism that can support nodes to dynamically join or exit during the participation process according to the predetermined security policy, and can effectively respond to changes in node attributes, ensuring data privacy. Secondly, this scheme combined blockchain technology and used its smart contract mechanism to record the operation content, achieving the openness and transparency of the system operation process and enhancing the security of attribute revocation. Through scheme analysis, this paper have proved that the ciphertext generated by the algorithm has good indistinguishability. The performance analysis also effectively demonstrates the advantages of this scheme.

Programmable network technology controls network devices and data packets through software-defined and programming techniques, enhancing network flexibility, scalability, and automation capabilities, thereby laying a solid foundation for the development of multimodal networks. Based on a programmable architecture, this paper designed a data packet routing and forwarding mechanism for six modalities: identity, content, geographical location, elastic address space, IPv4, and IPv6, and implemented packet parsing, routing lookup, and forwarding at the data plane. Simultaneously, a multimodal network control system was constructed to support functions such as packet parsing, topology management, flow table generation and distribution, and network measurement. It integrated resource coordination and scheduling algorithms to analyze network status in real time, compute routing rules, and distribute flow tables. Through traffic feature extraction, this paper achieves security detection and builds a multimodal network traffic time-series model based on deep learning to realize anomaly detection and identification, introducing intrinsic security features to ensure system availability and reliability. Experimental results demonstrate that the proposed scheme enables unified communication and control of multimodal networks, supporting multiple modalities. The control system is functionally complete and performs stably, with a topology scale exceeding 2000 nodes and end-to-end latency below 100ms. The security detection function can identify abnormal traffic and network modalities in real time, with an anomaly detection accuracy rate of 96.49% and a modality recognition accuracy rate of 99.72%.

To address the bottlenecks in signature data storage and verification efficiency in existing blockchain systems, this paper investigated multi-signature algorithms based on Schnorr and BLS. With the widespread adoption of blockchain applications, traditional ECDSA signature schemes face challenges such as high computational overhead and significant storage consumption when handling large volumes of signatures. To this end, the Schnorr multi-signature scheme significantly reduced resource consumption through its signature aggregation property, while the BLS scheme enhanced signature verification efficiency and storage efficiency via bilinear mapping. The integration of these multi-signature algorithms improved the overall performance of blockchain systems. Experimental results demonstrate that, compared to the traditional ECDSA scheme, the smart contract-based Schnorr and BLS multi-signature schemes exhibit clear advantages in security, computational overhead, and storage efficiency. Furthermore, the paper proposed a dynamic threshold m-of-n multi-signature scheme based on smart contracts, allowing real-time adjustment of signature policies (e.g., switching from 3-of-5 to 4-of-6) according to requirements. A multi-signature wallet dApp supporting multi-user collaborative asset management was also designed and implemented. System test results validate the security and usability of this wallet application, providing a practical foundation for blockchain multi-signature technology.

The article proposed a novel reversible neural network-based video steganography model. By replacing the widely used dense connection functions in current reversible neural networks with an optimized non-local mechanism function, the model effectively addressed the issue of low-quality secret information extraction from video carriers. Through the design of a superposition-based encryption and decryption structure, the model’s security was enhanced. Additionally, by optimizing the number of reversible neural network blocks, the network efficiency was improved. Experimental results demonstrate that, when utilizing reversible neural networks for video steganography, compared to dense network structures, the reversible neural network employing the nonlocal mechanism recovers secret information of higher quality, and the video distortion after secret embedding is reduced. Furthermore, the use of the superposition-based encryption and decryption structure significantly enhances the security of reversible neural network-based video steganography at the application level.

With the continuous evolution of web applications and widespread deployment of database-driven systems, SQL injection attacks remain a critical research focus in web security defense due to their highly covert and destructive nature. To address challenges posed by structural complexity, semantic diversity, and scarcity of attack samples in SQL injection attack statements, this paper proposed a novel detection method integrating graph structure modeling with reinforcement learning mechanisms. The proposed approach models SQL statements as graph structures and leverages an enhanced Graph Attention Network (GAT) to fuse syntactic features from nodes and edges. A multi-agent reinforcement learning framework incorporating four specialized detection experts was constructed to enable dynamic ensemble decision-making. Additionally, an adversarial sample generation module specifically designed for SQL injection obfuscation characteristics enhanced the model’s capability in identifying complex mutation attacks. Furthermore, explainability analysis using LIME and SHAP methods improved system transparency and practical applicability. Experimental results demonstrate that the proposed method effectively mitigates detection bias caused by sample imbalance and attack pattern diversification while maintaining low computational resource consumption. The method achieves 0.955 detection accuracy and 0.978 AUC on comprehensive SQL injection datasets, significantly outperforming existing baseline methods and providing an effective solution for intelligent SQL injection attack detection.