Large language models (LLMs) are typically equipped with alignment mechanisms designed to prevent the generation of harmful content. However, recent advances have led to the emergence of highly evasive disguised prompt jailbreak attacks (DPJAs), where attackers conceal real malicious intent within prompts that appear benign on the surface, thereby inducing the model to produce unsafe outputs. In this work, we propose prompt reconstruction-based detection (RePrompt), a training-free and plug-in detection framework designed to identify such jailbreak attacks. The key insight of RePrompt is that, when guided by carefully designed templates, LLMs possess the capability to uncover disguised adversarial prompts and effectively detect jailbreak attacks. RePrompt leverages the reasoning capabilities of the LLM itself to analyze and uncover the true intent behind a user’s query. When a disguised jailbreak prompt is encountered, RePrompt reconstructs the underlying malicious intent hidden beneath the surface-level disguise. Experiments across multiple LLMs, datasets, and three representative DPJAs demonstrate that RePrompt consistently outperforms state-of-the-art defense methods, reducing the average attack success rate from 47.0% to below 1.6%, achieving a near-zero false positive rate, and limiting the average query cost to only 1.13.

The challenge of Domain Generalization (DG) in Face Anti-Spoofing (FAS) is the significant interference of domain-specific signals on subtle spoofing clues. Recently, some CLIP-based algorithms have been developed to alleviate this interference by adjusting the weights of visual classifiers. How-ever, our analysis of this class-wise prompt engineering suffers from two shortcomings for DG FAS: (1) The categories of facial categories, such as real or spoof, have no semantics for the CLIP model, making it difficult to learn accurate category descriptions. (2) A single form of prompt cannot portray the various types of spoofing. In this work, instead of class-wise prompts, we propose a novel Content-aware Composite Prompt Engineering (CCPE) that generates instance-wise composite prompts, including both fixed template and learnable prompts. Specifically, our CCPE constructs content-aware prompts from two branches: (1) Inherent content prompt explicitly benefits from abundant transferred knowledge from the instruction-based Large Language Model (LLM). (2) Learnable content prompts implicitly extract the most informative visual content via Q-Former. Moreover, we design a Cross-Modal Guidance Module (CGM) that dynamically adjusts unimodal features for fusion to achieve better generalized FAS. Finally, our CCPE has been validated for its effectiveness in multiple cross-domain experiments and achieves state-of-the-art (SOTA) results.

Face anti-spoofing (FAS) based on domain generalization (DG) has attracted increasing attention from researchers.The reason for the poor generalization is that the model is overfitted to salient liveness-irrelevant signals.However, the previous methods alleviate the overfitting by mapping the images from multiple domains into a common feature space or promoting the separation of image features from domain-specific features and task-related features.If the text features of vision-language pre-trained (VLP) models (e.g., CLIP) are used to dynamically adjust the image features to gain a better generalization, we can not only explore a wider feature space but also avoid the potential degradation of semantic information.Specifically, we propose a FAS method of Style-Conditional Prompt Token Learning (S-CPTL), which aims to generate generalized text features by training the introduced prompt tokens to carry visual styles and use them as weights for classifiers to improve the model's generalization.Compared to the inherently static prompt token, we propose the dynamic prompt token, which can adaptively capture live-irrelevant signals from the instance-specific styles and increase their diversity through mixed feature statistics to further reduce the overfitting of the model.Thorough experimental analysis demonstrates that S-CPTL exceeds current top-performing methods in four distinct cross-dataset benchmarks.

Nowadays, the network-centric computing environment poses numerous security risks. Network Security Situation Awareness (NSSA) is a crucial action that involves monitoring and preventing ongoing and potential threats to information systems. To support this work, this paper proposes an extended cybersecurity ontology, which can be utilized as the core component of NSSA implementations. Our ontology consists of several modular sub-ontologies, each of which has a specific function that contributes to NSSA. And it is relatively more comprehensive and better organized than existing security ontologies. To demonstrate the applicability and efficiency of our ontology, we applied it to the construction process of a practical knowledge graph. We also conducted experiments in two practical situation awareness scenarios to validate the efficacy of our ontology in monitoring system situations. In the first case we apply our ontology to mine potential ongoing attack behaviors and discover attack scenarios. In the second validation scenario, we obtain the trend of changing situation based on the knowledge graph. The results of these experiments showed that our ontology can be easily used to link available information and perform effectively in practical situation awareness scenarios.

Dataset Distillation (DD), a newly emerging field, aims at generating much smaller but efficient synthetic training datasets from large ones. Existing DD methods based on gradient matching achieve leading performance; however, they are extremely computationally intensive as they require continuously optimizing a dataset among thousands of randomly initialized models. In this paper, we assume that training the synthetic data with diverse models leads to better generalization performance. Thus we propose two model augmentation techniques, i.e. using early-stage models and parameter perturbation to learn an informative synthetic set with significantly reduced training cost. Extensive experiments demonstrate that our method achieves up to 20× speedup and comparable performance on par with state-of-the-art methods.

Despite the advantages of scalability and flexibility, Security Function Virtualization (SFV) raises concerns about its own security. To enhance the security of SFV, a promising approach is to run critical components of off-the-shelf security software inside Software Guard Extensions (SGX) enclaves. This idea, however, is hardly practical due to the difficulty of detaching components from the monolithic security function and the unacceptable cost of executing them inside enclaves. In this article, we propose S-Blocks, an architecture to modularize virtual security functions (VSFs) and protect crucial modules with SGX in an efficient manner. S-Blocks decomposes VSFs into trusted and untrusted modules and provides dedicated APIs systematically. Only crucial VSF modules are hardened with enclaves. Furthermore, aiming at addressing state consistency and secure migration issues of security function scaling, we design a fine-grained state synchronization and migration mechanism to ensure loss-free, order-preserving, and state security for VSFs. To demonstrate the effectiveness of our approach, we prototype S-Blocks using Fast-Click on a real Skylake platform and implement three critical types of virtual security functions based on the S-Blocks architecture. Our evaluation results show that S-Blocks only imposes a manageable performance overhead, and low latency and resource consumption when protecting VSFs.

The popularization of 5G and the development of cloud computing further promote the application of images. The storage of images in an untrusted environment has a great risk of privacy leakage. This paper outlines a design for a lightweight image encryption algorithm based on a message-passing algorithm with a chaotic external message. The message-passing (MP) algorithm allows simple messages to be passed locally for the solution to a global problem, which causes the interaction among adjacent pixels without additional space cost. This chaotic system can generate high pseudorandom sequences with high speed performance. A two-dimensional logistic map is utilized as a pseudorandom sequence generator to yield the external message sets of edge pixels. The external message can affect edge pixels, and then adjacent pixels interact with each other to produce an encrypted image. A MATLAB simulation shows the cipher-image performs fairly uniform distribution and has acceptable information entropy of 7.996749. The proposed algorithm reduces correlation coefficients from plain-image 1 to its cipher-image 0, which covers all of the plain-image characters with high computational efficiency (speed = 18.200374 Mbit/s). Theoretical analyses and experimental results prove the proposed algorithm’s persistence to various existing attacks with low cost.

DDoS attacks have caused very serious damage to enterprise networks. Recently, a new kind of DDoS attack called link-flooding attack (LFA), has surfaced and is already being used by attackers to flood and congest network critical links. LFA is very difficult to detect since adversaries often utilize large-scale legitimate low-speed flows and rolls target links to isolate target areas for launching attacks. To address such a critical security problem, we design and implement a novel LFA defense system called LFADefender that leverages some key features, such as programmability, network-wide view, and flow traceability, of an emerging network technology, Software-Defined Networking (SDN), to effectively detect and migrate LFA. In LFADefender, we propose a LFA target link selection approach and design a LFA congestion monitoring mechanism to effectively detect LFA. In addition, we present a multiple optional paths rerouting method to temporarily mitigate links congestion caused by LFA. We further propose a malicious traffic blocking approach to radically mitigate LFA. Our evaluation results show that LFADefender can accurately detect and rapidly mitigate LFA, but only imposes minimal overhead in the communication channels between network controllers and data planes.

Moving target defense (MTD) has provided a dynamic and proactive network defense to reduce or move the attack surface that is available for exploitation. However, traditional network is difficult to realize dynamic and active security defense effectively and comprehensively. Software-defined networking (SDN) points out a brand-new path for building dynamic and proactive defense system. In this paper, we propose CHAOS, an SDN-based MTD system. Utilizing the programmability and flexibility of SDN, CHAOS obfuscates the attack surface including host mutation obfuscation, ports obfuscation, and obfuscation based on decoy servers, thereby enhancing the unpredictability of the networking environment. We propose the Chaos Tower Obfuscation (CTO) method, which uses the Chaos Tower Structure (CTS) to depict the hierarchy of all the hosts in an intranet and define expected connection and unexpected connection. Moreover, we develop fast CTO algorithms to achieve a different degree of obfuscation for the hosts in each layer. We design and implement CHAOS as an application of SDN controller. Our approach makes it very easy to realize moving target defense in networks. Our experimental results show that a network protected by CHAOS is capable of decreasing the percentage of information disclosure effectively to guarantee the normal flow of traffic.

Abstract Trusted Platform Module (TPM) is the “root of trust” of the whole trusted computing platform. The TPM's own security assurance is very important. This paper describes the TPM 2.0 hash‐based message authentication code (HMAC) authorization scheme as a security protocol and makes a detail comparison of the TPM 2.0 authorization to the TPM 1.2 “Object‐Independent Authorization Protocol” and the “Object‐Specific Authorization Protocol.” Then the authors use the typed pi calculus to describe the TPM 2.0 HMAC authorization and its security properties under the Digital Rights Management (DRM) scenario and use ProVerify to reason that the key handle manipulation attack for TPM 1.2 does not exist any more in TPM 2.0, because the access entity unique name has been linked to the HMAC value, but the vulnerability of key blob substitution still exists in TPM 2.0. Copyright © 2015 John Wiley & Sons, Ltd.

In this paper, a framework of end to end (E2E) trusted cloud infrastructure is proposed. On one end of the cloud provider, the trusted chain is extended to VMM and VM by trusted measurement and remote attestation, which can assure the trust of VMM and VM. On another end of the cloud terminal, the trusted mechanism is used to protect the terminal security. For the trust of cloud network, trusted network connect (TNC) is leveraged to protect the security of communication between the loud provider and the cloud terminal. The E2E trusted cloud infrastructure provides an E2E trusted protection for cloud computing. In addition, it can support the Chinese cryptographic algorithm (SMx) based on TPM 2.0.

The development of Internet, especially World Wide Web, brings new opportunities and challenges to integration of spatial information. The characteristics of spatial information sources over the Internet are distributed and heterogeneous. Integration of information across distributed, heterogeneous spatial information sources is still a challenge to both of the research community and the industry. This paper proposes a mediation and GML-based architecture for querying and integrating spatial information across multiple information sources, which recognizes the autonomous and heterogeneous nature of the current practice of Internet by adding GML-enabled wrappers for distributed servers and mediator tier. And a prototype based on the proposed architecture is presented in this paper