Ensuring the secure and proper deployment of Enclaves is critical for the security of Trusted Execution Environments (TEEs). Recently, the increasing attacks targeting Enclaves have made the traditional solution risky to use different enclaves for each task. An ideal solution is a pre-designed enclave that can be flexibly reused across different computational tasks. To address this issue, we propose FH-TEE, a novel framework that integrates Fully Homomorphic Encryption (FHE) with TEE. In this approach, computational tasks are partitioned so that only fixed, logic-independent operations are performed within the Enclave, enabling its reuse across various tasks. Flexibility in accommodating diverse computational requirements is achieved through FHE. Furthermore, because computationally intensive FHE operations, such as bootstrapping and ciphertext transformation, are logic-independent and offloaded to the Enclave, FH-TEE significantly improves efficiency compared to purely FHE-based solutions, while maintaining robust privacy and security. In our demo program, the server receives ciphertext from the client, performs ten iterations of the Game of Life computation, and then sends the resulting ciphertext back to the client. The results show that our scheme is 10.57× faster in computation speed and 32064× smaller in transmission size compared to a pure FHE scheme.
As artificial intelligence and deep learning models become increasingly sophisticated, processing vast amounts of personal data, the need for robust privacy protection measures has become paramount. This paper addresses the critical challenge of implementing privacy-preserving neural networks (PPNN) using fully homomorphic encryption (FHE), with a focus on the TFHE scheme. We present novel approaches to homomorphically implement maxpooling, a crucial non-linear operation in convolutional neural networks, which has been challenging to realize within existing FHE frameworks. Our main contributions include two innovative homomorphic maxpooling algorithms: HSMaxPool, utilizing the non-linear Sign function, and HRMaxPool, based on a homomorphic ReLU function. Both algorithms are implemented using TFHE’s gate bootstrapping and programmable bootstrapping techniques, respectively. Experiments were conducted on neural networks operating over ciphertext, employing the proposed homomorphic maxpooling algorithms. The results show that our methods deliver improved inference accuracy and reduced execution time compared to prior approaches, while maintaining accuracy throughout the maxpooling phase. This demonstrates the effectiveness of our homomorphic maxpooling algorithms in enhancing performance without sacrificing precision, further advancing the feasibility of efficient and privacy-preserving neural network computations on encrypted data.
A ring signature can be performed by anyone in a group of users who holds a signing key, and the signature of a message reveals nothing about the signer’s identity except the fact that they are a ring member. Currently, the size of most ring signatures depends on the number of ring members, but for realistic applications with a large number of ring members, this will lead to higher communication overheads. To reduce the signature size, in this paper, a new transparent compact ring signature scheme is constructed. We use homomorphic encryption and constrained signatures to improve the efficiency of the signature scheme when the ring size is large, thereby achieving “fully compact” signatures whose size is O(1), i.e., constant in the number of ring members. Not only does the scheme guarantee the privacy of the signer’s identity with less communication overhead, but it also provides stronger security under the RO model, including unforgeability and anonymity, under the assumption of the quantum hardness of the standard SIS problem. Furthermore, to the best of our knowledge, our scheme is the first compact post-quantum ring signature scheme that has transparent setup, requiring only public randomness (and not any trusted entity).
Identifying related Indicators of Compromise (IoCs), such as malicious IP addresses and associated network services (IP:Ports), is crucial for threat intelligence but challenging due to the scale and heterogeneity of Internet scan data. We propose a novel self-supervised contrastive learning framework to generate meaningful embeddings for IP:Port instances by fusing multimodal features (TLS certificates, banners, web content, metadata). Addressing the critical false negative problem inherent in contrastive learning, especially when combined with Hard Negative Mining (HNM), we introduce Semantic-Consistency HNM (SC-HNM). This technique employs a lightweight, integrated auxiliary network operating on early-stage feature representations to assess semantic similarity between an anchor and potential hard negatives. If the initial similarity exceeds a threshold δ , the potentially false negative is dynamically filtered out before computing the main contrastive loss, thus preserving semantic structure without requiring any labels. Our multimodal encoder utilizes attention and gating mechanisms for effective feature fusion. We train the model entirely from scratch on 8 million unlabeled scan records. Experiments show our approach yields highly competitive IoC retrieval performance, significantly outperforming zero-shot baselines (including large pre-trained models like BERT, RoBERTa, BGE-M3, E5). Ablation studies confirm the benefits of multimodality, HNM, and our SC-HNM filter. Furthermore, the resulting model is compact ( 27 million parameters) and efficient, enabling practical deployment for large-scale IoC similarity search and threat discovery.
Private Set Intersection (PSI) enables a sender and a receiver to jointly compute the intersection of their sets without disclosing non-trivial information about other items. However, depending on the context of joint data analysis, information derived from the items in the intersection may also be considered sensitive. To protect such sensitive information, prior work proposed Differentially private PSI (DPSI), which can be instantiated with circuit-PSI using Fully Homomorphic Encryption. Although asymptotically efficient, its concrete performance is sub-optimal compared with the practical state-of-the-art (SOTA) circuit-PSI. In this paper, we propose two generic DPSI constructions with provable security and privacy. We identify the critical criteria for selecting essential PSI-related tools by revisiting the DPSI definition. Then, we present two generic DPSI constructions. The first construction allows us to achieve provable privacy and efficiency by integrating any circuit-PSI with the randomized response mechanism. By plugging the SOTA circuit-PSI protocol, we obtain a DPSI protocol with concrete performance enhancement. The second construction offers a more efficient DPSI alternative by using multi-query Reverse Private Membership Test (mqRPMT) at the price of intersection size leakage. However, such leakages can be bounded with differential privacy by padding random dummy items in input sets. We conduct comprehensive experiments with various instantiations. The experiments show that our instantiations significantly outperform the existing DPSI construction: 2.9-26.8 × more communication-efficient and up to 110.5-151.8 × faster.
Non-asymptotic complexity estimates for attacks on Learning with Errors (LWE) problems are critically important. The Meet-LWE algorithm, proposed by May at CRYPTO 2021, presents an advanced attack method with non-asymptotic complexity estimates [1]. However, detailed discussions of its parameter selection is in lack. In this paper, we first investigate Meet-LWE, analyze specific parameter selection criteria, including the determination of search tree depths and the value of ϵ , a parameter employed to expand the representation space. Our results show that in practical attacks, ϵ is often constrained to remain within a limited range, preventing Meet-LWE from achieving its expected performance under certain parameters. Secondly, we follow the line of Wenger et al.’s work [2], characterize the real-world memory consumption of solving LWE with sparse secrets building upon May’s method. We have implemented a real-world attack against weakened Kyber, and achieved optimizations in both time and memory consumption. Compared to the hybrid dual lattice attack proposed in [2], we optimize memory consumption by approximately 15 times.
While many machine learning-based Network Intrusion Detection Systems (NIDS) have demonstrated near-perfect classification accuracy on static datasets, this “static perfection” often masks their inherent fragility when confronting continuously evolving, real-world cyber threats. When the evaluation paradigm shifts from static cross-validation to more realistic time-series scenarios, the performance of these models undergoes catastrophic degradation due to the presence of “concept drift.” To address the adaptability challenge of static models in dynamic environments, this paper proposes a novel dynamic dual-autoencoder framework—DyNA-IDS—capable of efficient online learning and self-adaptation in non-stationary data streams. The core of this framework is a unique, unsupervised drift detection module composed of a “benchmark autoencoder” (trained on benign traffic) and a “drift autoencoder” (trained on attack traffic) operating in parallel. Innovatively discarding traditional fixed-threshold methods, it instead utilizes a dual-stream parallel Kolmogorov-Smirnov (K-S) test mechanism. This approach continuously monitors the distribution of the reconstruction error difference between the two autoencoders across both long and short time scales, enabling the synchronous and highly sensitive perception of both gradual and sudden drifts. Once a drift is confirmed, a composite loss function, which includes an elastic regularization term, guides the entire framework through an incremental joint training process. This serves to effectively mitigate catastrophic forgetting while adapting to new attack patterns. Experimental results on a large-scale, public time-series network traffic dataset show that DyNA-IDS exhibits exceptional performance on dynamic data streams, achieving an average F1-score of 0.996 and significantly outperforming static models as well as multiple classic baseline drift detection methods. This research not only provides a robust, end-to-end framework for addressing concept drift in NIDS but also highlights the importance of evaluating models in dynamic environments, offering a solid theoretical and practical foundation for building truly resilient, next-generation cybersecurity systems. The source code is available at: https://github.com/GZHU-Innovation-Intersection-Lab/DyNA-IDS .
In the era of advanced data collection technologies, safeguarding individual privacy has become paramount. Local Differential Privacy (LDP) provides an effective means of privacy preservation, widely adopted across academia and industry. This paper investigates LDP perturbation mechanisms—specifically focusing on encoding, perturbation, and aggregation—under a sampling scheme without replacement, which differs from traditional approaches that utilize replacement-based sampling. Our work introduces refined variance analyses and proposes enhanced perturbation mechanisms tailored for scenarios involving entire populations. We find that the Improved Simmons mechanism achieves the lowest variance when the privacy budget is below a specific threshold, while the Improved Warner mechanism excels beyond this threshold. These results align closely with real-world contexts. In experiments using actual datasets, the Improved Warner mechanism reduces variance to 28.7 https://github.com/ssjhf/LDP-Perturbation-Mechanisms.git .
Ransomware has posed a significant threat to cybersecurity. Although techniques such as anomalous behavior detection have proven effective in mitigating its threats, several potential challenges have emerged that may undermine the reliability of these defenses. Specifically, the large volume of events generated during attacks causes significant challenges for efficient ransomware analysis, thereby hindering timely detection and response. Besides, malicious processes are interwoven with benign activities, which introduces irregular noise and compromises the effectiveness of detection. To address these challenges, we propose FuzzyHawk, which excels at extracting behavior patterns and detecting ransomware behaviors under noisy environments. First, it employs a graph-based representation of process behaviors and incorporates a high-frequency subgraph extraction mechanism to distill representative behavioral patterns efficiently. Besides, it employs a fuzzy matching mechanism that tolerates disturbances of system noise and enhances the robustness of ransomware detection. Experimental results on a large-scale public dataset (about 50 million events, covering 378 ransomware samples) demonstrate that FuzzyHawk achieves a remarkable true positive rate of 97.3
Deobfuscation techniques play a crucial role in enhancing the readability and analyzability of obfuscated code, thereby facilitating more effective reverse engineering and security analysis. However, the majority of existing deobfuscation research has primarily focused on the source code level, with limited effectiveness in deobfuscating binary code. Despite the emergence of Large Language Models (LLMs) and their effective application in the deobfuscation domain, there remains a lack of research leveraging LLMs for deobfuscating ARM binary code. In this paper, we construct a high-quality dataset of obfuscated ARM assembly code, comprising seven types of single obfuscations and seven types of multiple obfuscations. Based on this, we propose LLM-DAS, the first deobfuscation system specifically designed for ARM binary code and grounded in large language models. LLM-DAS consists of two finetuned large models tailored for obfuscation detection and deobfuscation, respectively, and we evaluate its performance. The experimental results demonstrate that the obfuscation detection component achieves an average accuracy and precision of 91.96
TWINE, a lightweight block cipher based on the Generalized Feistel Structure, was proposed by Suzaki et al. in SAC 2012. Since TWINE was proposed, its security has attracted considerable attention. However, to the best of our knowledge, no prior work has been published on related-key rectangle attacks against TWINE. Therefore, in this paper, we provide a comprehensive security evaluation of TWINE in the context of such attacks. Our approach focuses on the construction of key-recovery-friendly related-key boomerang distinguisher. Building upon the method of searching single-key boomerang distinguishers proposed by Hadipour et al., we model the precise differential characteristics (rather than the truncated differential characteristics) over the r_0 and r_1 rounds. Additionally, we model the propagation of both precise and truncated differential characteristics for the round function and key schedule and constrain the number of active S-boxes in the input and output of the boomerang distinguishers. As a result, we present the longest-known distinguishers for both TWINE-80 and TWINE-128. Using these distinguishers, we perform related-key rectangle attacks that improve upon previous results. Specifically, leveraging a 21-round boomerang distinguisher, we successfully mount a 25-round related-key rectangle attack on TWINE-80, improving upon the previous attack by one round. For TWINE-128, we conduct a 28-round related-key rectangle attack using a 23-round boomerang distinguisher, significantly reducing the time complexity required for the attack.
Multi-key Fully Homomorphic Encryption (MKFHE) enables computation between ciphertexts encrypted with different keys. Multi-key bootstrapping is the key techniques to construct TFHE-like MKFHE. In this paper, we proposed two efficient multi-key bootstrapping algorithms under NTRU and (R)LWE assumptions. They are based on a new multi-key blind rotation framework, which uses NTRU-based GSW-like encryption to perform single-key blind rotation and extends it to multi-key blind rotation by re-encryption and key-switching. Compared to the state of the art, our methods achieve significant gains in both runtime and key size efficiency. Concretely, for the number of keys k = 16 at the same security level, our method is 2.5-9.0 times faster than existing methods and reduces key size by a factor of up to 17.
The hash-based digital signature (HBS) algorithm is becoming essential in the field of digital signatures due to its quantum resistance. However, current research on HBS algorithms primarily focuses on the design of individual algorithms, lacking a comprehensive database that encompasses complete knowledge of HBS algorithms. This poses significant challenges for cross-scheme performance comparisons, and algorithm recommendations based on varying scenario requirements. To fill this blank, we propose an HBS algorithmic database construction method based on Chain-of-Thought (CoT) guidance, which innovatively adopts CoT and snowballing technologies to effectively enhance the completeness of the database and the consistency of algorithm relationships. Specifically, we first designed a cryptography-specific prompt template called Chain-of-Thought Knowledge Extraction (CoT-KE) to guide large models in feature inference, thereby accurately extracting algorithm-related content. Then we developed a systematic approach that combines the Snowballing with Relationship Type Identification and Enhancement (SRTIE) to construct a cross-literature knowledge network, tracking the derivative paths of algorithms and improving the integrity of algorithm knowledge. Moreover, we proposed a verification framework called Multi-View Quality Validation and Consistency (MV-QVC) to enhance the accuracy of the information extracted from algorithms. Extensive experiments demonstrate that our method achieves an accuracy of 94.3
Functional Encryption (FE) is a crucial cryptographic primitive that enables partial message recovery. In the static bounded model, the number of decryption key queries is fixed at the outset of the system. There have been many constructions in this model. However, it has certain limitations. Recently, Agrawal et al. (Crypto 2021) and Garg et al. (Eurocrypt 2022) introduced the dynamic bounded model, where the collusion bound is no longer fixed globally, but can be selected fluidly by each encryptor. Dynamic bounded FE can be seen as a balance between static bounded FE and normal FE, offering enhanced flexibility and adaptability in various cryptographic applications. Garg et al. (PKC 2024) proposed two general constructions for transitioning from static bounded FE to dynamic bounded FE using identity based encryption (IBE). One construction utilizes a binary tree of garbled circuits, while the other, which is referred to as ’the simpler construction’ henceforth, employs a single garbled circuit. While Garg et al. provided a high-level proof idea for the simpler construction, we present a detailed simulation-based security proof, demonstrating its validity even when the underlying IBE achieves only selective-id security. Furthermore, we demonstrate that in the indistinguishable-style security definition, dynamic bounded FE can be constructed from a selective-tag secure tagged static bounded FE, offering additional pathways for building dynamic bounded FE.
With the development of quantum computing technology, the threat of quantum computing to classical cryptography has been significantly enhanced. Studying the impact of quantum computing on the security of classical cryptosystems is a research focus in the cryptographic community. Pseudorandom functions are fundamental building blocks in numerous cryptosystems and play a critical role in post-quantum cryptography. To the best of our knowledge, research on effective quantum attacks on pseudorandom functions remains limited in the noisy intermediate-scale quantum era. This paper proposes a distributed quantum key recovery attack on pseudorandom functions and designs the corresponding quantum circuit based on distributed quantum computing technology, Bernstein-Vazirani’s and Grover’s algorithms. The proposed attack is applied to two pseudorandom functions, pEDM and SoKAC. Compared with the existing attacks, the proposed attack has lower complexity and stronger robustness to circuit noise. The proposed attack can realize effective quantum key recovery and provide theoretical support for designing quantum secure pseudorandom functions.
To address the challenges of high-dimensional traffic feature redundancy and the scarcity of novel attack samples in 5G networks and IoT environments, this paper proposes a multimodal and collaborative intelligent detection framework named SMCE-ICHBO-CBGAR. This study focuses on three critical technical challenges. First, we propose a Synthetic Multi-Cluster Equilibrium (SMCE) Approach, which overcomes the limitations caused by distribution shifts in traditional balancing methods in adversarial attack scenarios by combining K-means feature space decoupling with a hybrid SMOTE enhancement technique. Secondly, we construct an Improved Chaotic Honey Badger Optimization (ICHBO) based on tent chaotic initialization and parallel fitness evaluation to accelerate feature selection. In addition, in order to solve the shortcomings of traditional intrusion detection methods in robustness and complex attack pattern recognition, this paper designs a spatiotemporal attention residual fusion model (CBGAR) through hybrid feature encoding of CNN and BiGRU, a multi-head self-attention mechanism, and a multi-scale residual connection. Experiments conducted on the CICIDS-2017 and CICIoT 2023 datasets show that the accuracy of the proposed model reaches 97.60
Modern Controller Area Network (CAN) buses lack native security, leaving vehicles exposed to spoofing, replay, and injection attacks, especially zero-day or unseen variants that evade traditional IDSs. We present CANalyze-AI, an edge-optimized hybrid IDS combining Random Forest and XGBoost with a 4-bit, LoRA-adapted GPT-2 to add semantic reasoning under strict resource budgets. Upon flagging anomalous 50-frame windows, the LLM produces concise, human-readable rationales and drafts Sigma rules that pass schema checks before use. On a composite CAN dataset, CANalyze-AI completes detection-plus-explanation in under 100 ms per window, fits within a ≤ 4 GB RAM envelope, and improves F_1 by +0.9 ≥ 12
Enhanced Privacy ID (EPID) is a variant of group signature that allows a revocation manager to revoke signing keys and signatures. It maintains a corresponding key revocation list and signature revocation list in parallel, making it widely used for device attestation in real-world trusted execution environments and included in ISO/IEC standards. However, the EPID signature schemes currently applied in standards are vulnerable to attacks by quantum computers. In recent years, several post-quantum secure EPID schemes have been proposed. Notably, Chen et al. (PQCrypto 2024) introduced an efficient hash-based EPID scheme featuring an additional concrete non-interactive zero-knowledge proof to verify signer non-revocation. This proof has linear complexity relative to the signature revocation list size. We identify room for efficiency improvements of this proof in linear complexity and highlight the need for a more generalized approach to post-quantum secure EPID scheme construction. In this paper, we propose a general framework for constructing post-quantum secure EPID schemes based on the OR sigma protocol from admissible pair of group actions. We instantiate this framework using isogeny and lattice-based techniques. The core technique of our approach involves generating a revocation tokens list based on a randomized current signature revocation list and embedding its proof into the OR sigma protocol, eliminating the need to produce an independent proof. Compared to the most efficient post-quantum secure counterpart by Chen et al. (PQCrypto 2024) under the same group size, signature revocation list size, and at a higher security level (NIST-2 vs. NIST-1), our lattice-based EPID instantiation achieves a 4.33 × reduction in signature size while maintaining comparable signing and verification times. Additionally, our isogeny-based instantiation represents the first EPID scheme constructed using isogeny and achieves the smallest signature size among existing post-quantum secure counterparts. Furthermore, our security analysis demonstrates that the proposed scheme satisfies anonymity and unforgeability under the random oracle model.
The growing use of deep learning in intrusion detection systems (IDS) has increased the need for explainable IDS (X-IDS). Current X-IDS research mainly relies on local surrogate models and their explanations are limited due to the complex features of network traffic. In contrast, counterfactual explanations improve interpretability by contrasting “why P rather than Q”, avoiding complex absolute explanations. In this paper, we outline the theoretical value of counterfactual explanations in X-IDS, and propose a two-stage search method for generating counterfactuals. Moreover, we also propose a submodular pick method for counterfactual explanations that provides diverse explanation instances, aiming to interpret the entire model rather than individual samples. Experimental results demonstrate that our proposed method achieves an 80
