IoT malware is often generated in large quantities through source code modifications and widely used in cyberattacks. For fine-grained functional analysis, it is crucial to extract and utilize relevant behavioral information from the executables of malware samples. In this study, our primary objective is to improve the clustering accuracy of IoT malware based on phylogenetic trees, and from the results, we investigate which components are useful for analyzing functional characteristics at a level more fine-grained than the family.Executable code can generally be classified into statically linked library functions and user-written code, the latter presumed to reflect attacker-specific functionality. Based on this fact, we hypothesized that meaningful functional distinctions among samples are primarily captured in the user code regions. To test this hypothesis, we constructed multiple datasets with and without user code and library functions, and evaluated clustering quality with respect to two functional label sets: disabling information and exploit information.The experimental results showed that, as hypothesized, excluding library functions and retaining only the opcodes of user code improves clustering accuracy for disabling information, whereas for exploit labels, accuracy improves even when only the library function regions are used. These results demonstrate that eliminating irrelevant information is effective for functional label classification, and that functional analysis does not necessarily rely solely on user code; rather, it was revealed that vulnerability-related information is also contained in the library parts.
We present a CKKS native homomorphic transciphering framework that converts existing LEA-128 ciphertexts into LEA-256 ciphertexts without exposing plaintext. Motivated by the post-quantum reduction in symmetric key security, our design shows that, by using homomorphic encryption, especially CKKS, it is possible to transition to post-quantum cryptography without revealing sensitive information. Furthermore, we propose a method that optimizes ARX operations by leveraging the SIMD structure of CKKS. Our approach combines: (i) bitwise packing that maps LEA’s ARX to CKKS primitive operations; (ii) a CKKS adapted Kogge–Stone adder that implements modular addition with logarithmic carry depth using masked rotations; and (iii) a low degree noise attenuation polynomial that preserves binary fixed points and suppresses near binary perturbations at a modest depth overhead. With 2^15 CKKS slots and 32 CPU threads, the pipeline that converts LEA-128 to LEA-256 completes with an amortized time of 6.61 s, and the results decrypt correctly under our parameters. Compared with TFHE style Boolean gate baselines, our circuit formulation and 1-bit encoding deliver over 3 × speedups for bootstrapped modular addition and subtraction, while XOR and rotation are substantially faster due to SIMD packing. These results indicate that CKKS based transciphering is practical for ARX ciphers like LEA and provides a confidential migration path from 128-bit to 256-bit keys without plaintext exposure in the post-quantum era.
At ACISP 2017, Naito proposed a double-block-length (DBL) hash function with the following features: (i) without the costly feed-forward operations; (ii) of rate 1/2, and every pair of blockcipher-calls is parallelizable; (iii) achieving n-bit indifferentiability security using a blockcipher with n-bit blocks and κ -bit keys. This is one of the state-of-the-art DBL hash constructions. The only shortage is the use of a finalization function by two blockcipher-calls, which is unfriendly to short messages. We consider , which is obtained by injecting Naito’s feed-forward-free compression function into the Merkle-Damgård with permutation construction (JoC 2008), and prove n - log _2 n indifferentiability for it. To our knowledge, this is the first time to achieve the three aforementioned properties of plus finalization-freeness simultaneously. We also investigate the (second) preimage security of and . When κ≥ 2n+1 and a strengthened Merkle-Damgård message padding is used, for any challenge message with at least n+2 blocks, we show an attack finding a second preimage on with the same number of blocks, using O(n 2^n) time and O( 2^n ) memory. Our idea also yields a preimage attack against with the same complexities.
LINE is the most widely used messaging application worldwide, with 196 million monthly active users as of March 2025. Given its broad adoption, ensuring the security of its communication protocol is of significant importance. However, its end-to-end encryption protocol, Letter Sealing, has not undergone comprehensive formal verification. Although Letter Sealing Version 2—introduced after vulnerabilities were identified in Version 1—is now widely deployed, its security guarantees remain insufficiently scrutinized. In this work, we formally model the one-to-one communication protocol in both Letter Sealing versions using the symbolic verification tool Tamarin-Prover and analyze key security properties, including confidentiality, authentication, and integrity. For Letter Sealing Version 1, we reproduce previously reported vulnerabilities, such as impersonation and replay attacks, and additionally conduct an exhaustive verification of a broad range of security properties. For Letter Sealing Version 2, we represent the first comprehensive formal analysis. We confirm that several vulnerabilities have been mitigated through changes in cryptographic algorithms; however, our analysis also reveals several previously unidentified weaknesses that persist in the protocol. This work provides a foundational step toward a rigorous evaluation of the security guarantees offered by LINE’s encryption mechanisms and underscores the ongoing need for formal verification in widely deployed communication systems.
ChaCha is an ARX-based stream cipher and has been widely used in practice. In this paper, we propose a new derivation rule for linear approximations of ARX-based ciphers and then apply it to ChaCha. Compared with the previous derivation rules for linear approximations, it can reduce the number of active bits when deriving linear approximations. By combining this new derivation rule with the previous derivation rules in a mixed pattern, we improve the 3.5- and 4-round linear approximations for ChaCha. The original 3.5-round linear approximation presented at EUROCRYPT 2021 has a correlation of 2^-47 , which was improved to be -2^-41 recently. This paper further improves it with a better correlation of -2^-40 . Besides, we enhance the 4-round linear approximation for ChaCha with correlation 2^ - 48 , improving the previous result (i.e., -2^-50 ) by a factor of -2^2 . The new derivation rule for linear approximations is general and can be applied to other ARX-based ciphers.
We evaluate the power side-channel leakage of the lattice-based KEM SMAUG-T, focusing on the polynomial-to-message conversion. Under identical STM32F4 conditions, leakage metrics significantly exceed standard thresholds: TVLA |t|=41.07 , SNR=8.96 , and CPA correlation |ρ |=0.934 . The root cause is the sequential byte-wise construction with repeated STRB stores, which produces a cumulative Hamming-weight “staircase” signature. A parallel evaluation of ML-KEM under identical experimental conditions confirms the same vulnerability pattern, with peak SNR=8.21 and |ρ |=0.91 , showing that this is a systematic issue rather than scheme-specific. Our practical attack exploits this leakage using a lightweight MLP classifier, achieving 95.8
Modern web services utilize complex architectures involving diverse HTTP processing components. These components include CDNs, WAFs, proxies, and web servers (e.g., Nginx, Apache, Cloudflare, and GCP). Independent parsing rules across these components lead to discrepancies in interpreting identical HTTP requests, including the URI. This phenomenon, known as Path Confusion, leads to critical security vulnerabilities, such as authentication bypass and sensitive data exposure. However, existing penetration testing methods fail to identify these vulnerabilities adequately. This limitation stems from limited system-wide visibility and operational risks in live environments. To address these challenges, we present PathFault, an automated methodology leveraging a Predicate Logic-based Surrogate Model. This model systematically abstracts component-specific parsing behaviors, enabling precise vulnerability analysis with minimal invasive testing. Furthermore, PathFault incorporates SMT solving techniques to automatically generate targeted exploit payloads. It combines security researchers’ domain expertise with web service-specific parsing logic. Consequently, our approach significantly improves detection accuracy. It comprehensively assesses dynamic system interactions and addresses limitations of traditional root-cause analysis. Evaluation demonstrates PathFault’s efficacy in identifying previously overlooked Path Confusion vulnerabilities.
Vulnerability detectors based on language models (LMs) show promise, but their robustness under adversarial conditions remains underexplored. Specifically, prior evaluations often lack diverse adversarial attack strategies and sophisticated perturbation implementations. To address this, we present AdvCodeGen, a black-box adversarial framework that leverages a large language model (LLM) to systematically evaluate the robustness of various LM-based vulnerability detectors. Unlike purely Abstract syntax tree (AST)-based rule transformations, AdvCodeGen utilizes an LLM to not only implement 19 effective code perturbation rules but also to generate more diverse and syntactically varied adversarial examples while preserving functionality. We evaluated AdvCodeGen against six state-of-the-art LMs (CodeT5, CodeBERT, GraphCodeBERT, UnixCoder, StarCoder2, and CodeGen2.5) on the BigVul and Devign datasets. AdvCodeGen achieves attack success rates (ASR) up to 81.5
As software complexity increases and cyberattacks grow more sophisticated, the need for tools that can automatically identify threats within systems has become increasingly evident. However, prior research has primarily focused on detecting attacks or vulnerabilities at individual points within a system, limiting its ability to identify multi-step threats that arise when adversaries exploit the order in which data is transferred among system components. To address this limitation, this paper extends the Data Flow Diagram (DFD) metamodel of the OWASP-ODTM project by incorporating additional information such as data-transfer sequence. Using this extended metamodel, we developed a methodology capable of automatically identifying multi-step threats within a system. We further implemented this methodology in pytm, an existing threat-modelling automation tool, and proposed an enhanced version named pytm-Seq. To demonstrate that pytm-Seq can identify multi-step threats exploited in real-world attacks, we analyzed 52 incidents and derived 7 multi-step threat patterns. We then verify whether these patterns can be detected within DFDs using pytm-Seq. Finally, we quantitatively evaluate the tool’s detection accuracy for the patterns and observe an accuracy of 86 https://github.com/sinse100/pytm-Seq ).
Witness Encryption (WE) enables encryption under computational problems where decryption requires knowledge of a valid witness to a nondeterministic polynomial-time (NP) statement. While theoretical constructions exist for all NP languages using complex assumptions like multilinear maps, practical schemes remain limited to specialized relations. We present the first extractable Witness Key Encapsulation Mechanism (WKEM) for Quadratic Arithmetic Programs (QAPs), the mathematical foundation underlying modern zkSNARKs including Groth16. Our construction transforms Groth16’s verification equations into an encryption mechanism, enabling practical witness encryption for rich algebraic relations expressible as arithmetic circuits. We prove extractability in the Algebraic Group Model combined with the Random Oracle Model, providing a clean reduction to Groth16’s computational knowledge soundness without requiring novel assumptions. Our WKEM composes generically with symmetric encryption to yield a complete extractable WE scheme for QAP relations. We validate our construction through a complete implementation with practical performance: circuits with 162–150K constraints achieve encryption/decryption in ∼20 ms to ∼1.7 s. Our approach bridges the gap between WE theory and practice by building on the mature zkSNARK ecosystem while maintaining strong extractability guarantees.
The emergence of CPU-GPU integrated System-on-Chip architectures like Apple’s M1 has made Unified Memory Architecture mainstream in computing. While UMA eliminated the CPU-GPU data copy overhead that was a major bottleneck in traditional dGPU systems, it shifted the optimization focus to kernel-internal memory access patterns. This paper analyzes how memory optimization strategies based on data access patterns affect cryptographic algorithm performance in SoC environments. We implemented the block cipher ARIA in Counter (CTR) mode on Apple M1 GPU using the Metal framework, designing six implementation configurations that systematically vary the storage locations of ARIA’s primary data structures—S-boxes and round keys—across device, threadgroup, and constant memory spaces. Experimental results confirmed that memory strategies optimized for data access patterns have a decisive impact on performance. A hybrid strategy placing S-boxes with irregular access patterns in threadgroup memory and round keys with uniform broadcast access patterns in constant memory achieved 451.93 MB/s throughput, representing 47.07
Modern military operations are increasingly evolving into Network-Centric Warfare (NCW), where the security of tactical communications has become a critical concern. Traditional cryptographic methods, while ensuring confidentiality, inadvertently reveal the presence of sensitive messages, making them potential targets for detection and analysis. To address this limitation, this study proposes a novel steganographic approach that conceals the existence of tactical messages by embedding them within natural language text, thereby achieving both secrecy and plausible deniability. Leveraging recent advances in generative AI, particularly the LLaMA-2 language model, we develop a text-based steganography system tailored for military communication. A custom dataset was constructed by combining the UCDP conflict event records with the literary corpus of Treasure Island, enabling domain adaptation through fine-tuning with QLoRA. The resulting system demonstrates strong performance across multiple evaluation metrics, achieving a message reconstruction accuracy of 90.70
The quantum resistance of lattice-based cryptography becomes moot if its deployment introduces exploitable side-channels. We present the first comprehensive vulnerability assessment of NIST standardized post-quantum algorithms within ARM TrustZone. Through systematic evaluation on NXP i.MX93 hardware running OP-TEE 4.4, we demonstrate that integrating Dilithium-3 and Kyber-768 into existing TEE architectures creates critical vulnerabilities: (i) near-deterministic operation classification (98.9 × world-switch frequency versus classical algorithms; (ii) rejection sampling leaking 0.192 bits per signing operation, enabling key recovery within 50,000 observations; (iii) complete denial-of-service where Dilithium’s 73KB footprint exceeds the 32KB TA heap; and (iv) 40 × latency increase under concurrent workloads. Our analysis traces these vulnerabilities to fundamental TEE design decisions—synchronous invocation, shared caches, static memory pools—incompatible with PQC computational patterns. We quantify mitigation efficacy and release our measurement framework for reproducible assessment.
The Generalized Feistel Network (GFN) has long served as a versatile foundation for lightweight block cipher design, with the even-odd shuffle variant offering particularly strong diffusion properties. Recently, MGFN was introduced under Malaysia’s National Cryptography Policy, followed by MGFN-P, which replaces the original S-box with that of PRESENT to enhance resistance against differential attacks. In this work, we present a comparative study of MGFN, MGFN-P, and TWINE to reassess the root causes of MGFN’s cryptanalytic weaknesses. Our analysis shows that these weaknesses cannot be attributed solely to the low branch number of MGFN’s S-box, but arise primarily from the structural permutations used in its design. To investigate whether security can be improved through alternative permutations while fixing the S-box, we evaluate MGFN variants under both nibble-wise and bit-wise permutations. By exploiting equivalence relations, we reduce 40320 nibble-wise candidates to only 22 equivalence classes and identify variants with stronger resistance than MGFN and MGFN-P, though still weaker than TWINE. In the bit-wise setting, exploratory experiments with cyclic shifts and random permutations yield stronger variants, but none match TWINE’s robustness. Overall, our results underscore the limitations of MGFN’s design and reaffirm the reliability of well-established GFN configurations for lightweight cryptography.
This paper presents vector-input keyed hashing modes that construct collision-resistant pseudorandom functions (CR PRFs) using a keyed hash function, where a vector refers to a sequence of variable-length strings. The proposed vector-input keyed hashing modes, VIM1 and VIM2, originate from the intuition that a string-input keyed hashing mode using a compression function results in a vector-input keyed hashing mode by replacing the compression function with a hash function. Combined with the recently proposed string-input keyed hashing modes KHC1 or KHC2, VIM1 and VIM2 are shown to yield CR PRFs from a compression function satisfying extended collision resistance and being a secure PRF under related-key attacks. Extended collision resistance means that it is intractable to find a distinct input pair whose output difference falls within a small set. This paper also introduces a keyed hashing mode, PVIM, which allows parallel processing of strings in a vector. However, it requires more calls to the underlying keyed hash function than VIM1 and VIM2 do. To the best of our knowledge, this is the first proposal of dedicated vector-input CR PRFs.
In recent years, the rapid evolution and complexity of cyber-attacks have made robust intrusion detection systems (IDSs), particularly network-based IDSs (NIDSs), a critical component of cybersecurity infrastructures. Although traditional machine learning (ML) and deep learning (DL) approaches have been widely explored, they often face challenges related to scalability, generalisation, and efficiency. Transformers, originally developed for natural language processing, show a strong ability to face the challenges of ML and DL models. The research that studies applying transformers for detecting intrusions, especially considering their performance on different data representations (tabular vs. image-transformed), is still weak and limited. Therefore, this paper investigates the potential of transformer-based architectures to enhance intrusion detection performance. We conduct a comprehensive evaluation of several transformer models, including TabNet, SAINT-Transformer, FT-Transformer, and Vision Transformer (ViT), across multiple intrusion detection datasets: NSL-KDD, UNSW-NB15, and UNR-IDD. TabNet, SAINT-Transformer, and FT-Transformer are evaluated exclusively on tabular data, while Vision Transformer (ViT) and a lightweight transformer are evaluated on both tabular and image-based representations of network traffic. Image representations for ViT are generated using two methods: single-sample transformation and the sliding window technique. Our experiments assess model performance using accuracy, precision, recall, and F1-score. The results show that tabular transformers perform strongly on datasets with high feature complexity, while ViT with sliding window images excels on datasets with strong temporal dependencies, achieving performance comparable to hybrid models such as CNN+LSTM. These findings provide valuable insights into selecting appropriate transformer architectures and data representations for building efficient and effective NIDS solutions.
As post-quantum cryptography becomes the cornerstone of future security systems, it is crucial to understand its practical resilience under present side-channel analysis. In this paper, we propose an effective side-channel analysis method for the poly_tomsg() function in the decapsulation of the first-order mask implementation of CRYSTALS-Kyber (the NIST-selected post-quantum key encapsulation mechanism). Unlike previous methods that target individual bits, our method targets a masked shared byte. By systematically evaluating multiple neural network architectures (MLP, CNN, ResNet), we find that residual learning has the best robustness under low signal-to-noise ratios. The method achieves up to 98
This paper examines the resistance of Rijndael-256 to related-key attacks by presenting robust boomerang and rectangle distinguishers. We first revisit the local collision construction method used by Biryukov et al. to find related-key differential trails for AES-192 and AES-256. Leveraging that approach, we show that local collisions cannot be constructed by separating disturbance and correction in Rijndael-256 by analyzing how differences propagate through its key schedule and their implications for related-key trails. In particular, the internal state and the key state are updated simultaneously, which makes it impossible to construct a related-key differential distinguisher based on local collisions. Our in-depth analysis yields 7-round related-key boomerang and rectangle distinguishers for Rijndael-256 without relying on local collisions. With data complexities of 2^108.83 and 2^167.92 256-bit blocks, respectively, these distinguishers enable us to distinguish Rijndael-256 from a random oracle in the related-key setting. To the best of our knowledge, this is the first analysis of related-key boomerang and rectangle attacks on Rijndael-256 in the literature.
The Fujisaki-Okamoto (FO) transformation is a method of practically realizing CCA-secure public-key encryption (PKE) scheme. Recently, the FO transformation for PKE in PQC has also been widely studied. In this work, we focus on how to practically realize CCA security in identity-based encryption (IBE) in PQC. Our contributions are as follows: (i) We show that a slightly modified post-quantum FO transformation can be effectively used to transform from CPA-secure IBE scheme to CCA-secure IBE scheme in the quantum random oracle model (QROM). (ii) Our main contribution is to show that the FO transformation realizes anonymity of IBE scheme in addition to CCA-security if the underlying CPA-secure IBE scheme meets the additional property called strong disjoint-simulatable (SDS) property. (iii) To demonstrate that the above result (ii) is not mere theory, we also show that a CPA-secure IBE scheme with the SDS property can indeed be constructed based on the Gentry-Peikert-Vaikuntanathan (GPV) IBE scheme. Consequently, combining (ii) and (iii) yields a post-quantum anonymous IBE scheme with CCA security in the QROM.
Efficient implementation of pairing-based cryptosystems relies on high-performance arithmetic in finite fields 𝔽_p and their extensions 𝔽_p^k , where k denotes the embedding degree. A small embedding degree is crucial since a part of the arithmetic for pairing computation occurs in 𝔽_p^k , including squaring, multiplication, and Frobenius operations. In this paper, we present a fast and efficient method for computing the Frobenius endomorphism in the finite field 𝔽_p^k , along with an analysis of its computational complexity. Furthermore, we introduce an optimization for cyclotomic cubing operations tailored to several pairing-friendly elliptic curves, which are essential for the calculation of Tate pairing and its derivatives.
