In this work, we explore neural network design options for discriminating Random Number Generators(RNG), as a complement to existing statistical test suites, being a continuation of a recent paper of the aothors. Specifically, we consider variations in architecture and data preprocessing. We test their impact on the network's ability to discriminate sequences from a low-quality RNG versus a high-quality one—that is, to discriminate between "optimal" sequence sets and those from the generator under test. When the network fails to distinguish them, the test is passed. For this test to be useful, the network must have real discrimination capabilities. We review several network design possibilities showing significant differences in the obtained results. The best option presented here is convolutional networks working on 5120-byte sequences.

At ASIACRYPT’19, Bonnetain et al. demonstrated that an S-box can be distinguished from a permutation chosen uniformly at random by quantifying the distances between their behaviors. In this study, we extend this approach by proposing a deep learning-based method to quantify distances between two different S-boxes and evaluate similarities in their design structures. First, we introduce a deep learning-based framework that trains a neural network model to recover the design structure of a given S-box based on its cryptographic table. We then interpret the decision-making process of our trained model to analyze which coefficients in the table play significant roles in identifying S-box structures. Additionally, we investigate the inference results of our model across various scenarios to evaluate its generalization capabilities. Building upon these insights, we propose a novel approach to quantify distances between structurally different S-boxes. Our method effectively assesses structural similarities by embedding S-boxes using the deep learning model and measuring the distances between their embedding vectors. Furthermore, experimental results confirm that this approach is also applicable to structures that the model has never seen during training. Our findings demonstrate that deep learning can reveal the underlying structural similarities between S-boxes, highlighting its potential as a powerful tool for S-box reverse-engineering.

In this work, we introduce ToFA, the first fault attack (FA) strategy that attempts to leverage the classically well-known idea of impossible differential cryptanalysis to mount practically verifiable attacks on bit-oriented ciphers like GIFT and BAKSHEESH. The idea used stems from the fact that truncated differential paths induced due to fault injection in certain intermediate rounds of the ciphers lead to active SBox-es in subsequent rounds whose inputs admit specific truncated differences. This leads to a (multi-round) impossible differential distinguisher, which can be incrementally leveraged for key-guess elimination via partial decryption. The key-space reduction further exploits the multi-round impossibility, capitalizing on the relations due to the quotient-remainder (QR) groups of the GIFT and BAKSHEESH linear layer, which increases the filtering capability of the distinguisher. Moreover, the primary observations made in this work are independent of the actual SBox. Clock glitch based fault attacks were mounted on 8-bit implementations of GIFT-64/GIFT-128 using a ChipWhisperer Lite board on an 8-bit ATXmega128D4-AU micro-controller. Unique key recovery was achieved for GIFT-128 with 3 random byte faults, while for GIFT-64, key space was reduced to $2^{32}$, the highest achievable for GIFT-64, with a single level fault due to its key-schedule. This work also reports the highest fault injection penetration for any variant of GIFT and BAKSHEESH. Finally, this work reiterates the role of classical cryptanalysis strategies in fault vulnerability assessment while leading to the most efficient fault attacks on GIFT.

Impossible differential attack is one of the major cryptanalytical methods for symmetric-key block ciphers. In this paper, we evaluate the security of SAND-128 against impossible differential attack. SAND is an AND-RX-based lightweight block cipher proposed by Chen et al. in Designs, Codes and Cryptography 2022. There are two variants of SAND, namely SAND-64 and SAND-128, due to structural differences. In this paper, we search for impossible differential distinguishers of SAND-128 using the Constraint Programming (CP) and reveal 14-round impossible differential distinguishers. The number of 14-round distinguishers is $2^{14} \times 7 = 114,688$. Furthermore, we demonstrate a key recovery attack on 21-round SAND-128. The complexities for the attack require $2^{124}$ data, $2^{127.2}$ encryptions, and $2^{122}$ bytes of memory, respectively. Although this result currently achieves the best attack on round-reduced SAND-128, this attack does not threaten the security of SAND-128 against impossible differential attack.

The paper analyzes the security of two recently proposed code-based cryptosystems that employ encryption of the form $y = m G_{\text{pub}} + eE_{pub}$: the Krouk-Kabatiansky-Tavernier (KKT) cryptosystem and the Lau-Ivanov-Ariffin-Chin-Yap (LIACY) cryptosystem. We demonstrate that the KKT cryptosystem can be reduced to a variant of the McEliece scheme, where a small set of columns in the public generator matrix is replaced with random ones. This reduction implies that the KKT cryptosystem is vulnerable to existing attacks on Wieschebrink's encryption scheme, particularly when Generalized Reed-Solomon (GRS) codes are used. In addition, we present a full key-recovery attack on the LIACY cryptosystem by exploiting its linear-algebraic structure and leveraging distinguishers of subcodes of GRS codes. Our findings reveal critical vulnerabilities in both systems, effectively compromising their security despite their novel designs.

The Boomerang attack was one of the first attempts to visualize a cipher ($E$) as a composition of two sub-ciphers ($E_1\circ E_0$) to devise and exploit two high-probability (say $p,q$) shorter trails instead of relying on a single low probability (say $s$) longer trail for differential cryptanalysis. The attack generally works whenever $p^2 \cdot q^2 > s$. However, it was later succeeded by the so-called ``sandwich attack'' which essentially splits the cipher in three parts $E'_1\circ E_m \circ E'_0$ adding an additional middle layer ($E_m$) with distinguishing probability of $p^2\cdot r\cdot q^2$. It is primarily the generalization of a body of research in this direction that investigate what is referred to as the switching activity and capture the dependencies and potential incompatibilities of the layers that the middle layer separates. This work revisits the philosophy of the sandwich attack over multiple rounds for NLFSR-based block ciphers and introduces a new method to find high probability boomerang distinguishers. The approach formalizes boomerang attacks using only ladder, And switches. The cipher is treated as $E = E_1 \circ E_m$, a specialized form of a sandwich attack which we called as the ``open-sandwich attack''. The distinguishing probability for this attack configuration is $r \cdot q^2$. Using this innovative approach, the study successfully identifies a deterministic boomerang distinguisher for the keyed permutation of the TinyJambu cipher over 320 rounds. Additionally, a 640-round boomerang with a probability of $2^{-22}$ is presented with 95% success rate. In the related-key setting, we unveil full-round boomerangs with probabilities of $2^{-19}$, $2^{-18}$, and $2^{-12}$ for all three variants, demonstrating a 99% success rate. Similarly, for Katan-32, a more effective related-key boomerang spanning 140 rounds with a probability of $2^{-15}$ is uncovered with 70% success rate. Further, in the single-key setting, a 84-round boomerang with probability $2^{-30}$ found with success rate of 60%. This research deepens the understanding of boomerang attacks, enhancing the toolkit for cryptanalysts to develop efficient and impactful attacks on NLFSR-based block ciphers.

Encrypted search schemes have been proposed to address growing privacy concerns. However, several leakage-abuse attacks have highlighted some security vulnerabilities. Recent attacks assumed an attacker's knowledge containing data "similar" to the indexed data. However, this vague assumption is barely discussed in literature: how likely is it for an attacker to obtain a "similar enough" data? Our paper provides novel statistical tools usable on any attack in this setting to analyze its sensitivity to data similarity. First, we introduce a mathematical model based on statistical estimators to analytically understand the attackers' knowledge and the notion of similarity. Second, we conceive statistical tools to model the influence of the similarity on the attack accuracy. We apply our tools on three existing attacks to answer questions such as: is similarity the only factor influencing accuracy of a given attack? Third, we show that the enforcement of a maximum index size can make the ``similar-data'' assumption harder to satisfy. In particular, we propose a statistical method to estimate an appropriate maximum size for a given attack and dataset. For the best known attack on the Enron dataset, a maximum index size of 200 guarantees (with high probability) the attack accuracy to be below 5%.

We study recent algebraic attacks (Briaud-Øygarden EC'23) on the Regular Syndrome Decoding (RSD) problem and the assumptions underlying the correctness of their attacks' complexity estimates. By relating these assumptions to interesting algebraic-combinatorial problems, we prove that they do not hold in full generality. However, we show that they are (asymptotically) true for most parameter sets, supporting the soundness of algebraic attacks on RSD. Further, we prove—without any heuristics or assumptions—that RSD can be broken in polynomial time whenever the number of error blocks times the square of the size of error blocks is larger than 2 times the square of the dimension of the code. Additionally, we use our methodology to attack a variant of the Learning With Errors problem where each error term lies in a fixed set of constant size. We prove that this problem can be broken in polynomial time, given a sufficient number of samples. This result improves on the seminal work by Arora and Ge (ICALP'11), as the attack's time complexity is independent of the LWE modulus.

This paper revisits the Hamming Weight (HW) labelling function for machine learning assisted side channel attacks. Contrary to what has been suggested by previous works, our investigation shows that, when paired with modern deep learning architectures, appropriate pre-processing and normalization techniques; it can perform as well as the popular identity labelling functions and sometimes even beat it. In fact, we hereby introduce a new machine learning method, dubbed, that helps solve the class imbalance problem associated to HW, while significantly improving the performance of unprofiled attacks. We additionally release our new, easy to use python package that we used in our experiments, implementing a broad variety of machine learning driven side channel attacks as open source, along with a new dataset AES_nRF, acquired on the nRF52840 SoC.

The ECDSA (Elliptic Curve Digital Signature Algorithm) is used in many blockchain networks for digital signatures. This includes the Bitcoin and the Ethereum blockchains. While it has good performance levels and as strong current security, it should be handled with care. This care typically relates to the usage of the nonce value which is used to create the signature. This paper outlines the methods that can be used to break ECDSA signatures, including revealed nonces, weak nonce choice, nonce reuse, two keys and shared nonces, and fault attack.

Screaming-channel attacks enable Electromagnetic (EM) Side-Channel Attacks (SCAs) at larger distances due to higher EM leakage energies than traditional SCAs, relaxing the requirement of close access to the victim. This attack can be mounted on devices integrating Radio Frequency (RF) modules on the same die as digital circuits, where the RF can unintentionally capture, modulate, amplify, and transmit the leakage along with legitimate signals. Leakage results from digital switching activity, so the hypothesis of previous works was that this leakage would appear at multiples of the digital clock frequency, i.e., harmonics. This work demonstrates that compromising signals appear not only at the harmonics and that leakage at non-harmonics can be exploited for successful attacks. Indeed, the transformations undergone by the leaked signal are complex due to propagation effects through the substrate and power and ground planes, so the leakage also appears at other frequencies. We first propose two methodologies to locate frequencies that contain leakage and demonstrate that it appears at non-harmonic frequencies. Then, our experimental results show that screaming-channel attacks at non-harmonic frequencies can be as successful as at harmonics when retrieving a 16-byte AES key. As the RF spectrum is polluted by interfering signals, we run experiments and show successful attacks in a more realistic, noisy environment where harmonic frequencies are contaminated by multi-path fading and interference. These attacks at non-harmonic frequencies increase the attack surface by providing attackers with an increased number of potential frequencies where attacks can succeed.

This survey, mostly written in the years 2022-2023, is meant as an as short as possible description of the current state-of-the-art lattice attacks on lattice-based cryptosystems, without losing the essence of the matter. The main focus is the security of the NIST finalists and alternatives that are based on lattices, namely CRYSTALS-Kyber, CRYSTALS-Dilithium and Falcon. Instead of going through these cryptosystems case by case, this survey considers attacks on the underlying hardness assumptions: in the case of the mentioned lattice-based schemes, these are (variants of) LWE (Learning With Errors) and NTRU.

SAND is an AND-RX-based lightweight block cipher proposed by Chen et al. There are two variants of SAND, namely SAND-64 and SAND-128, due to structural differences. In this paper, we search for impossible differential distinguishers of SAND-64 using the Constraint Programming (CP) and reveal 56 types of impossible differential distinguishers up to 11 rounds. Furthermore, we demonstrate a key recovery attack on 17-round SAND-64. The complexities for the attack require $2^{56}$ data, $2^{127}$ encryptions, and $2^{60}$ bytes of memory, respectively. Although this result currently achieves the best attack on round-reduced SAND-64, this attack does not threaten the security of SAND-64 against impossible differential attack.

Side-channel attacks consist of retrieving internal data from a victim system by analyzing its leakage, which usually requires proximity to the victim in the range of a few millimetres. Screaming channels are EM side channels transmitted at a distance of a few meters. They appear on mixed-signal devices integrating an RF module on the same silicon die as the digital part. Consequently, the side channels are modulated by legitimate RF signal carriers and appear at the harmonics of the digital clock frequency. While initial works have only considered collecting leakage at these harmonics, late work has demonstrated that the leakage is also present at frequencies other than these harmonics. This result significantly increases the number of available frequencies to perform a screaming-channel attack, which can be convenient in an environment where multiple harmonics are polluted. This work studies how this diversity of frequencies carrying leakage can be used to improve attack performance. We first study how to combine multiple frequencies. Second, we demonstrate that frequency combination can improve attack performance and evaluate this improvement according to the performance of the combined frequencies. Finally, we demonstrate the interest of frequency combination in attacks at $15$ and, for the first time to the best of our knowledge, at $30$ meters. One last important observation is that this frequency combination divides by $2$ the number of traces needed to reach a given attack performance.

We show that the attribute-based signature scheme [Information Sciences, 654(2024), 119839] is insecure, because an adversary can generate valid signatures for any message even though he cannot access the signer's secret key. The four components of signature $\{\delta_1, \delta_2, \delta_3, \delta_4\}$ are not tightly bound to the target message $M$ and the signer's public key. The dependency between the signer's public key and secret key is not properly used to construct any intractable problem. The inherent flaw results in that the adversary can find an efficient signing algorithm functionally equivalent to the valid signing algorithm.

In CRYPTO 2022, Esser et al. proposed a partial key exposure attack on several post-quantum cryptographic schemes including Rainbow which is a variant of UOV. The task of the attack is to recover a full secret key from its partial information such as a secret key with symmetric/asymmetric bit errors. One of the techniques Esser et al. developed is a partial enumeration that combines the standard algorithms to solve the MQ problem with enumeration. Although an efficient attack on Rainbow was proposed, UOV and its variants have still been paid much attention since UOV and its three variants, i.e., MAYO, QR-UOV and SNOVA, were selected as the Round 2 candidates of the additional call for digital signature schemes proposal by NIST. In this paper, we analyze partial key exposure attacks on UOV, MAYO, and QR-UOV. Although our proposed attacks use the partial enumeration, we refine their enumeration strategy. We employ two enumeration strategies and analyze the complexity of the proposed attacks. Then, we find a structural difference between UOV and its variants to resist partial enumeration. Specifically, the partial enumeration is effective if the number of vinegar variables is smaller than the number of equations and the order of a finite field is small. As a result, the proposed attack is the most effective on MAYO. While our attacks on UOV and QR-UOV are effective only when the symmetric error probabilities are 0.11 and 0.05, respectively, that on MAYO is effective even when the probability is close to 0.5.

In this article, we present for the first time a cross-core Prime+Probe attack on ARM TrustZone, which bypasses the AutoLock mechanism. We introduce our simulation- driven methodology based on gem5 for vulnerability analysis. We demonstrate its utility in reverse engineering a SoC platform in order to study its microarchitectural behavior (caches, etc.), inside a simulator, in spite of hardware protection. We present a novel vulnerability analysis technique, which takes into account the cache set occupancy for targeted victim executable. This proves to be essential in identifying information leakage in presence of AutoLock. The above tool also identifies the cache lines leaking a maximum amount of information. A cross-core Prime+Probe attack is then mounted on these max-leakage cache lines both in simulation for fine-tuning, and in real hardware. We validate our analysis and attack method on OP-TEE, an open-source trusted execution environment running on RockPi4 a board based on RK3399 SoC. More specifically we target the RSA subroutine in the MbedTLS library used inside OP-TEE. Despite the presence of AutoLock, multiplier obfuscation, and assuming a cross-core attack, we are able to retrieve 30% of the key bits, which can later be used in Branch-and-Prune methods to recover the full key.

The Sum of Even-Mansour (SoEM) construction, proposed by Chen et al. at Crypto 2019, has become the basis for designing some symmetric schemes, such as the nonce-based MAC scheme $\text{nEHtM}_{p}$ and the nonce-based encryption scheme $\text{CENCPP}^{\ast}$. In this paper, we make the first attempt to study the quantum security of SoEM under the Q1 model where the targeted encryption oracle can only respond to classical queries rather than quantum ones. Firstly, we propose a quantum key recovery attack on SoEM21 with a time complexity of $\tilde{O}(2^{n/3})$ along with $O(2^{n/3})$ online classical queries. Compared with the current best classical result which requires $O(2^{2n/3})$, our method offers a quadratic time speedup while maintaining the same number of queries. The time complexity of our attack is less than that observed for quantum exhaustive search by a factor of $2^{n/6}$. We further propose classical and quantum key recovery attacks on the generalized SoEMs1 construction (consisting of $s\geq 2$ independent public permutations), revealing that the application of quantum algorithms can provide a quadratic acceleration over the pure classical methods. Our results also imply that the quantum security of SoEM21 cannot be strengthened merely by increasing the number of permutations.

At CRYPTO 2015, Kirchner and Fouque claimed that a carefully tuned variant of the Blum-Kalai-Wasserman (BKW) algorithm (JACM 2003) should solve the Learning with Errors problem (LWE) in slightly subexponential time for modulus $q=\mathrm{poly}(n)$ and narrow error distribution, when given enough LWE samples. Taking a modular view, one may regard BKW as a combination of Wagner's algorithm (CRYPTO 2002), run over the corresponding dual problem, and the Aharonov-Regev distinguisher (JACM 2005). Hence the subexponential Wagner step alone should be of interest for solving this dual problem - namely, the Short Integer Solution problem (SIS) - but this appears to be undocumented so far. We re-interpret this Wagner step as walking backward through a chain of projected lattices, zigzagging through some auxiliary superlattices. We further randomize the bucketing step using Gaussian randomized rounding to exploit the powerful discrete Gaussian machinery. This approach avoids sample amplification and turns Wagner's algorithm into an approximate discrete Gaussian sampler for $q$-ary lattices. For an SIS lattice with $n$ equations modulo $q$, this algorithm runs in subexponential time $\exp(O(n/\log \log n))$ to reach a Gaussian width parameter $s = q/\mathrm{polylog}(n)$ only requiring $m = n + \omega(n/\log \log n)$ many SIS variables. This directly provides a provable algorithm for solving the Short Integer Solution problem in the infinity norm ($\mathrm{SIS}^\infty$) for norm bounds $\beta = q/\mathrm{polylog}(n)$. This variant of SIS underlies the security of the NIST post-quantum cryptography standard Dilithium. Despite its subexponential complexity, Wagner's algorithm does not appear to threaten Dilithium's concrete security.

Fruit-F is a lightweight short-state stream cipher designed by Ghafari et al. The authors designed this version of the cipher, after earlier versions of the cipher viz. Fruit 80/v2 succumbed to correlation attacks. The primary motivation behind this design seemed to be preventing correlation attacks. Fruit-F has a Grain-like structure with two state registers of size 50 bits each. In addition, the cipher uses an 80-bit secret key and an 80-bit IV. The authors use a complex key-derivation function to update the non-linear register which prevents the same key-bit alignment across fixed-length window of keystream bits, which is essentially what stops the correlation attacks. In this paper, we first present two attacks against Fruit-F. The first attack stems from the fact that the key-derivation can be rewritten as the Boolean xor of two key-dependent terms one of which is the Boolean OR of two bits of the key. Using this we show that the cipher does not offer 80-bit security: the effective key space of Fruit-F is slightly less than $2^{80}$, i.e. a simple brute force attack costs around $2^{80}-2^{49}$ time. The second is a differential attack using the cipher's complex initialization process. We show that under some given conditions, it is possible to have two initial vectors $V_1$ and $V_2$ that produce identical keystream vectors with any given key. Using this as a distinguisher, it is possible to collect enough linear and quadratic equations of the secret key to find it in practical time with very few keystream bits.

We demonstrate an attack on the soundness of a widely known optimization of the Gemini multilinear Polynomial Commitment Scheme (PCS). The attack allows a malicious prover to falsely claim that a multilinear polynomial takes a value of their choice, for any input point. We stress that the original Gemini multilinear PCS and HyperKZG, an adaptation of Gemini, are not affected by the attack.

In this paper, we propose the Differential Fault Attack (DFA) on three Homomorphic Encryption (HE) friendly stream ciphers \textsf{Masta}, \textsf{Pasta}, and \textsf{Elisabeth}. Both \textsf{Masta} and \textsf{Pasta} are \textsf{Rasta}-like ciphers with publicly derived and pseudorandom affine layers. The design of \textsf{Elisabeth} is an extension of \textsf{FLIP} and \textsf{FiLIP}, following the group filter permutator paradigm. All these three ciphers operate on elements over $\mathbb{Z}_p$ or $\mathbb{Z}_{2^n}$, rather than $\mathbb{Z}_2$. We can recover the secret keys of all the targeted ciphers through DFA. In particular, for \textsf{Elisabeth}, we present a new method to determine the filtering path, which is vital to make the attack practical. Our attacks on various instances of \textsf{Masta} are practical and require only one block of keystream and a single word-based fault. By injecting three word-based faults, we can theoretically mount DFA on two instances of \textsf{Pasta}, \textsf{Pasta}-3 and \textsf{Pasta}-4. For \textsf{Elisabeth}-4, the only instance of the \textsf{Elisabeth} family, we present two DFAs in which we inject four bit-based faults or a single word-based fault. With 15000 normal and faulty keystream words, the DFA on \textsf{Elisabeth}-4 can be completed in just a few minutes.

We show that the certificateless authentication scheme [Mob. Networks Appl. 2022, 27, 346-356] fails to keep anonymity, not as claimed. The scheme neglects the basic requirement for bit-wise XOR, and tries to encrypt data by the operator. The negligence results in some trivial equalities. The adversary can retrieve the user's identity from one captured string via the open channel.

Deep learning-based side-channel analysis has become a popular and powerful option for side-channel attacks in recent years. One of the main directions that the side-channel community explores is how to design efficient architectures that can break the targets with as little as possible attack traces, but also how to consistently build such architectures. In this work, we explore the usage of the JumpReLU activation function, which was designed to improve the robustness of neural networks. Intuitively speaking, improving the robustness seems a natural requirement for side-channel analysis, as hiding countermeasures could be considered adversarial attacks. In our experiments, we explore three strategies: 1) exchanging the activation functions with JumpReLU at the inference phase, training common side-channel architectures with JumpReLU, and 3) conducting hyperparameter search with JumpReLU as the activation function. While the first two options do not yield improvements in results (but also do not show worse performance), the third option brings advantages, especially considering the number of neural networks that break the target. As such, we conclude that using JumpReLU is a good option to improve the stability of attack results.

In this paper, we present a side-channel attack on the hardware AES accelerator of a Bluetooth chip used in millions of devices worldwide, ranging from wearables and smart home products to industrial IoT. The attack leverages information about AES computations unintentionally transmitted by the chip together with RF signals to recover the encryption key. Unlike traditional side-channel attacks that rely on power or near-field electromagnetic emissions as sources of information, RF-based attacks leave no evidence of tampering, as they do not require package removal, chip decapsulation, or additional soldered components. However, side-channel emissions extracted from RF signals are considerably weaker and noisier, necessitating more traces for key recovery. The presented profiled machine learning-assisted attack can recover the full encryption key from 90,000 traces captured at a one-meter distance from the target device, with each trace being an average of 10,000 samples per encryption. This is a twofold improvement over the correlation analysis-based attack on the same AES accelerator.

In recent years, the integration of deep learning with differential cryptanalysis has led to differential neural cryptanalysis, enabling efficient data-driven security evaluation of modern cryptographic algorithms. Compared to traditional differential cryptanalysis, differential neural cryptanalysis enhances the efficiency and automation of the analysis by training neural networks to automatically extract statistical features from ciphertext pairs. As research advances, neural distinguisher construction faces challenges due to the absence of a unified framework capable of cross-algorithm generalization and feature optimization. There's no systematic way to build a framework from data formats and network architectures, which limits their scalability across diverse ciphers and and their suitability for combining different cryptanalysis methods. While neural network training is data-driven, we lack interpretable explanations for the quality of differentially generated datasets. Therefore, there is an urgent need to combine cryptographic theory with data analysis methods to systematically evaluate dataset quality. This paper proposes a novel framework for constructing related-key neural differential distinguishers that integrates three core innovations: (1) multi-ciphertext multi-difference formats to enhance dataset diversity and feature coverage, (2) structural filtering for prioritizing high-probability differential paths aligned with cryptographic architectures, and (3) Deep Residual Shrinkage Network (DRSN) with adaptive thresholding to suppress noise and amplify critical differential features. By applying this framework to two standardized algorithms DES and PRESENT, our results demonstrate significant advancements. For DES, the framework achieves an 8-round related-key neural distinguisher and improves 6/7-round distinguisher accuracy by over 40%. For PRESENT, we construct the first 9-round related-key neural distinguisher, which outperforms existing neutral distinguishers in both round coverage and accuracy. Additionally, we employ kernel principal component analysis (KPCA) and K-means clustering to evaluate the quality of differential datasets for DES and PRESENT, revealing that clustering compactness strongly correlates with distinguisher performance. Furthermore, we propose a validation algorithm to verify differential combinations with cryptographic advantages from a machine learning perspective, identifying ‘good’ plaintext-key differential combinations. We apply this approach to the SIMECK algorithm, demonstrating its broad applicability. These findings validate the framework’s effectiveness in bridging cryptographic analysis with data-driven feature extraction and offer new insights for automated security evaluation of block ciphers.

In parallel with the standardization of lattice-based cryptosystems, the research community in Post-quantum Cryptography focused on non-lattice-based hard problems for constructing public-key cryptographic primitives. The Linear Code Equivalence (LCE) Problem has gained attention regarding its practical applications and cryptanalysis. Recent advancements, including the LESS signature scheme and its candidacy in the NIST standardization for additional signatures, supported LCE as a foundation for post-quantum cryptographic primitives. However, recent cryptanalytic results have revealed vulnerabilities in LCE-based constructions when multiple related public keys are available for one specific code rate. In this work, we generalize the proposed attacks to cover all code rates. We show that the complexity of recovering the private key from multiple public keys is significantly reduced for any code rate scenario. Thus, we advise against constructing specific cryptographic primitives using LCE.

This paper presents a security analysis of the South Korean Format-Preserving Encryption (FPE) standards FEA-1 and FEA-2. In 2023, Chauhan \textit{et al.} presented the first third-party analysis of FEA-1 and FEA-2 against the square attack. The authors proposed new distinguishing attacks covering up to three rounds of FEA-1 and five rounds of FEA-2, with a data complexity of $2^8$ plaintexts. Additionally, using these distinguishers, they presented key recovery attacks for four rounds of FEA-1 and six rounds of FEA-2, for 192-bit and 256-bit key sizes. The complexities of both the four-round FEA-1 and six-round FEA-2 key recovery attacks are $2^{137.6}$. \\ In this work, we successfully extend the number of rounds attacked for both FEA-1 and FEA-2, using the square attack technique. Specifically, we present a four-round distinguishing attack against FEA-1 and six-round distinguishing attack against FEA-2. The data complexities of these distinguishers are $2^{64}$ plaintexts. Furthermore, we apply these distinguishers to perform key recovery attacks on five rounds of FEA-1 and seven rounds of FEA-2, targeting the 256-bit key size. The time complexities of the presented key recovery attacks are $2^{193.6}$.

Systems such as file backup services often use content-defined chunking (CDC) algorithms, especially those based on rolling hash techniques, to split files into chunks in a way that allows for data deduplication. These chunking algorithms often depend on per-user parameters in an attempt to avoid leaking information about the data being stored. We present attacks to extract these chunking parameters and discuss protocol-agnostic attacks and loss of security once the parameters are breached (including when these parameters are not setup at all, which is often available as an option). Our parameter-extraction attacks themselves are protocol-specific but their ideas are generalizable to many potential CDC schemes.

Power side-channel (PSC) vulnerabilities present formidable challenges to the security of ubiquitous microelectronic devices in mission-critical infrastructure. Existing side-channel assessment techniques mostly focus on post-silicon stages by analyzing power profiles of fabricated devices, suffering from low flexibility and prohibitively high cost while deploying security countermeasures. While pre-silicon PSC assessments offer flexibility and low cost, the true nature of the power signatures cannot be fully captured through RTL or gate-level design. Although physical design-level analysis provides precise power traces, collecting data is time and resource-consuming at the layout level. To address this challenge, we propose, for the first time, a fast and efficient physical design-level PSC assessment framework using a graph neural network (GNN). This framework predicts dynamic power traces for new layouts, using them to assess physical design security through metrics evaluation. Our experiments on AES-GF layout implementations achieve a tremendous 133 times speedup compared to conventional simulation-based flow without sacrificing substantial accuracy.

SNOVA is one of the submissions in the NIST Round 1 Additional Signature of the Post-Quantum Signature Competition. SNOVA is a UOV variant that uses the noncommutative-ring technique to educe the size of the public key. SNOVA's public key size and signature size are well-balanced and have good performance. Recently, Beullens proposed a forgery attack against SNOVA, pointing out that the parameters of SNOVA can be attacked. Beullens also argued that with some slight adjustments his attacks can be prevented. In this note, we explain Beullens' forgery attack and show that the attack can be invalid by two different approaches. Finally, we show that these two approaches do not increase the sizes of the public keys or signatures and the current parameters satisfy the security requirement of NIST.

Distinguishing Goppa codes or alternant codes from generic linear codes [FGO+11] has been shown to be a first step before being able to attack McEliece cryptosystem based on those codes [BMT24]. Whereas the distinguisher of [FGO+11] is only able to distinguish Goppa codes or alternant codes of rate very close to 1, in [CMT23a] a much more powerful (and more general) distinguisher was proposed. It is based on computing the Hilbert series $\{\mathrm{HF}(d),~d\in \mathbb{N}\}$ of a Pfaffian modeling. The distinguisher of [FGO+11] can be interpreted as computing $\mathrm{HF}(1)$. Computing $\mathrm{HF}(2)$ still gives a polynomial time distinguisher for alternant or Goppa codes and is apparently able to distinguish Goppa or alternant codes in a much broader regime of rates as the one of [FGO+11]. However, the scope of this distinguisher was unclear. We give here a formula for $\mathrm{HF}(2)$ corresponding to generic alternant codes when the field size $q$ satisfies $q \geq r$, where r is the degree of the alternant code. We also show that this expression for$\mathrm{HF}(2)$ provides a lower bound in general. The value of $\mathrm{HF}(2)$ corresponding to random linear codes is known and this yields a precise description of the new regime of rates that can be distinguished by this new method. This shows that the new distinguisher improves significantly upon the one given in [FGO+11].

We show that the anonymous authentication and key establishment scheme [IEEE TDSC, 20(4), 3535-3545, 2023] fails to keep user anonymity, not as claimed. We also suggest a method to fix it.

Due to their simple security assessments, permutation-based pseudo-random functions (PRFs) have become widely used in cryptography. It has been shown that PRFs using a single $n$-bit permutation achieve $n/2$ bits of security, while those using two permutation calls provide $2n/3$ bits of security in the classical setting. This paper studies the security of permutation-based PRFs within the Q1 model, where attackers are restricted to classical queries and offline quantum computations. We present improved quantum-time/classical-data tradeoffs compared with the previous attacks. Specifically, under the same assumptions/hardware as Grover's exhaustive search attack, i.e. the offline Simon algorithm, we can recover keys in quantum time $\tilde{O}(2^{n/3})$, with $O(2^{n/3})$ classical queries and $O(n^2)$ qubits. Furthermore, we enhance previous superposition attacks by reducing the data complexity from exponential to polynomial, while maintaining the same time complexity. This implies that permutation-based PRFs become vulnerable when adversaries have access to quantum computing resources. It is pointed out that the above quantum attack can be used to quite a few cryptography, including PDMMAC and pEDM, as well as general instantiations like XopEM, EDMEM, EDMDEM, and others.

Many lattice-based crypstosystems employ ideal lattices for high efficiency. However, the additional algebraic structure of ideal lattices usually makes us worry about the security, and it is widely believed that the algebraic structure will help us solve the hard problems in ideal lattices more efficiently. In this paper, we study the additional algebraic structure of ideal lattices further and find that a given ideal lattice in a polynomial ring can be embedded as an ideal into infinitely many different polynomial rings by the coefficient embedding. We design an algorithm to verify whether a given full-rank lattice in $\mathbb{Z}^n$ is an ideal lattice and output all the polynomial rings that the given lattice can be embedded into as an ideal with bit operations $\mathcal{O}(n^3(\log n + B)^2(\log n)^2)$, where $n$ is the dimension of the lattice and $B$ is the upper bound of the bit length of the entries of the input lattice basis. We would like to point out that Ding and Lindner proposed an algorithm for identifying ideal lattices and outputting a single polynomial ring of which the input lattice can be regarded as an ideal with bit operations $\mathcal{O}(n^5B^2)$ in 2007. However, we find a flaw in Ding and Lindner's algorithm, and it causes some ideal lattices can't be identified by their algorithm.

Profiling side-channel analysis (SCA) is widely used to evaluate the security of cryptographic implementations under worst-case attack scenarios. This method assumes a strong adversary with a fully controlled device clone, known as a profiling device, with full access to the internal state of the target algorithm, including the mask shares. However, acquiring such a profiling device in the real world is challenging, as secure products enforce strong life cycle protection, particularly on devices that allow the user partial (e.g., debug mode) or full (e.g., test mode) control. This enforcement restricts access to profiling devices, significantly reducing the effectiveness of profiling SCA. To address this limitation, this paper introduces a novel framework that allows an attacker to create and learn from their own white-box reference design without needing privileged access on the profiling device. Specifically, the attacker first implements the target algorithm on a different type of device with full control. Since this device is a white box to the attacker, they can access all internal states and mask shares. A novel conditional generative adversarial network (CGAN) framework is then introduced to mimic the feature extraction procedure from the reference device and transfer this experience to extract high-order leakages from the target device. These extracted features then serve as inputs for profiled SCA. Experiments show that our approach significantly enhances the efficacy of black-box profiling SCA, matching or potentially exceeding the results of worst-case security evaluations. Compared with conventional profiling SCA, which has strict requirements on the profiling device, our framework relaxes this threat model and, thus, can be better adapted to real-world attacks.

Side-channel analysis (SCA) is a growing field in hardware security where adversaries extract secret information from embedded devices by measuring physical observables like power consumption and electromagnetic emanation. SCA is a security assessment method used by governmental labs, standardization bodies, and researchers, where testing is not just limited to standardized cryptographic circuits, but it is expanded to AI accelerators, Post Quantum circuits, systems, etc. Despite its importance, SCA is performed on an ad hoc basis in the sense that its flow is not systematically optimized and unified among labs. As a result, the current solutions do not account for fair comparisons between analyses. Furthermore, neglecting the need for interoperability between datasets and SCA metric computation increases students’ barriers to entry. To address this, we introduce SCAPEgoat, a Python-based SCA library with three key modules devoted to defining file format, capturing interfaces, and metric calculation. The custom file framework organizes side-channel traces using JSON for metadata, offering a hierarchical structure similar to HDF5 commonly applied in SCA, but more flexible and human-readable. The metadata can be queried with regular expressions, a feature unavailable in HDF5. Secondly, we incorporate memory-efficient SCA metric computations, which allow using our functions on resource-restricted machines. This is accomplished by partitioning datasets and leveraging statistics-based optimizations on the metrics. In doing so, SCAPEgoat makes the SCA more accessible to newcomers so that they can learn techniques and conduct experiments faster and with the possibility to expand on in the future.

Recently, Differential Fault Attacks (DFAs) have proven highly effective against stream ciphers designed for Hybrid Homomorphic Encryption (HHE). In this work, we present a table-based DFA framework called the \textit{shortcut attack}, which generalizes the attack proposed by Wang and Tang on the cipher \textsf{Elisabeth}. The framework applies to a broad sub-family of ciphers following the Group Filter Permutator (GFP) paradigm and enhances previous DFAs by improving both the fault identification and path generation steps. Notably, the shortcut attack circumvents the issue of function representation, allowing successful attacks even when the cipher's filter function cannot be represented over the ring it is defined on. Additionally, we provide complexity estimates for the framework and apply the shortcut attack to \textsf{Elisabeth-4} and its patches. As a result, we optimize the DFA on \textsf{Elisabeth-4}, requiring fewer keystreams and running faster than previous methods. Specifically, we achieve a DFA that requires only $3000$ keystreams, which is one-fifth of the previous best result. We also successfully mount a practical DFA on \textsf{Gabriel-4} and provide a theoretical DFA for \textsf{Elisabeth-b4}. For the latest patch, \textsf{Margrethe-18-4}, which follows the more general Mixed Filter Permutator (MFP) paradigm, we present a DFA in a stronger model. To the best of our knowledge, these are the first DFA results on the patches of \textsf{Elisabeth-4}. Finally, we derive security margins to prevent shortcut attacks on a broad sub-family of MFP ciphers, which can serve as parameter recommendations for designers.

We show that the aggregate signature scheme [IEEE Syst. J., 2023, 17(3), 3822-3833] is insecure against forgery attack. This flaw is due to that the ephemeral key or ephemeral value chosen in the signing phase is not indeed bound to the final signature. An adversary can sign any message while the verifier cannot find the fraud. We also suggest a revising method to frustrate this attack.

Hardware obfuscation is an active trustworthy design technique targeting threats in the IC supply chain, such as IP piracy and overproduction. Recent research on Intellectual Property (IP) protection technologies suggests that using embedded reconfigurable components (e.g., eFPGA redaction) could be a promising approach to hide the functional and structural information of security-critical designs. However, such techniques suffer from almost prohibitive overhead in terms of area, power, delay, and testability. This paper proposes an obfuscation technique called EvoLUTe+, which is a unique and more fine-grained redaction approach using smaller reconfigurable components (e.g., Look-Up Tables (LUTs)). EvoLUTe+ achieves fine-grained partitioning, sub-circuit coloring, and scoring of IP, and then encrypts the original IP through the substitution of some sub-circuits. Different attacks are used to test the robustness of EvoLUTe+, including structural and machine learning attacks, as well as Bounded Model Checking (BMC) attacks. The overhead of the obfuscation design is also analyzed. Experimental results demonstrate that EvoLUTe+ exhibits robustness with acceptable overhead while resisting such threat models.

We show that the threshold signature scheme [J. Ind. Inf. Integr. 39: 100593 (2024)] is insecure against forgery attack. An adversary can find an efficient signing algorithm functionally equivalent to the valid signing algorithm, so as to convert the legitimate signature $(sig, s, r_x)$ of message $m$ into a valid signature $(sig, s, r_x')$ of any message $m'$.

In 2020, Castryck-Decru-Smith constructed a hash function using the (2,2)-isogeny graph of superspecial principally polarized abelian surfaces. In their construction, the initial surface was chosen from vertices quite "close" to the square of a supersingular elliptic curve with a known endomorphism ring. In this paper, we propose an algorithm for recovering a collision on their hash function. Under some heuristic assumptions, the time complexity and space complexity of our algorithm are estimated to be $\widetilde{O}(p^{3/10})$ which is smaller than the complexity $\widetilde{O}(p^{3/2})$ the authors had claimed necessary to recover such a collision, where $p$ is the characteristic of the base field. In particular case where $p$ has a special form, then both the time and space complexities of our algorithm are polynomial in $\log{p}$. We implemented our algorithm in MAGMA, and succeeded in recovering a collision in 17 hours (using 64 parallel computations) under a parameter setting the authors had claimed to be 384-bit secure. Finally, we propose a simple countermeasure against our attack, which is expected to restore the complexity required to recover a collision to $\widetilde{O}(p)$ currently.

This study explores the algebraic cryptanalysis of small-scale variants of the E0 stream cipher, a legacy cipher used in the Bluetooth protocol. By systematically reducing the size of the linear feedback shift registers (LFSRs) while preserving the cipher’s core structure, we investigate the relationship between the number of unknowns and the number of consecutive keystream bits required to recover the internal states of the LFSRs. Our work demonstrates an approximately linear relationship between the number of consecutive keystream bits and the size of small-scale E0 variants, as indicated by our experimental results. To this end, we utilize two approaches: the computation of Gröbner bases using Magma’s F4 algorithm and the application of CryptoMiniSat’s SAT solver. Our experimental results show that increasing the number of keystream bits significantly improves computational efficiency, with the F4 algorithm achieving a speedup of up to 733× when additional equations are supplied. Furthermore, we verify the non-existence of equations of degree four or lower for up to seven consecutive keystream bits, and the non-existence of equations of degree three or lower for up to eight consecutive keystream bits, extending prior results on the algebraic properties of E0.

The key collision attack was proposed as an open problem in key-committing security in Authenticated Encryption (AE) schemes like $\texttt{AES-GCM}$ and $\texttt{ChaCha20Poly1305}$. In ASIACRYPT 2024, Taiyama et al. introduce a novel type of key collision—target-plaintext key collision ($\texttt{TPKC}$) for $\texttt{AES}$. Depending on whether the plaintext is fixed, $\texttt{TPKC}$ can be divided into $\texttt{fixed-TPKC}$ and $\texttt{free-TPKC}$, which can be directly converted into collision attacks and semi-free-start collision attacks on the Davies-Meyer ($\texttt{DM}$) hashing mode. In this paper, we propose a new rebound attack framework leveraging a time-memory tradeoff strategy, enabling practical key collision attacks with optimized complexity. We also present an improved automatic method for finding \textit{rebound-friendly} differential characteristics by controlling the probabilities in the inbound and outbound phases, allowing the identified characteristics to be directly used in $\textit{rebound-based}$ key collision attacks. Through our analysis, we demonstrate that the 2-round $\texttt{AES-128}$ $\texttt{fixed-TPKC}$ attack proposed by Taiyama et al. is a $\texttt{free-TPKC}$ attack in fact, while $\texttt{fixed-TPKC}$ attacks are considerably more challenging than $\texttt{free-TPKC}$ attacks. By integrating our improved automatic method with a new rebound attack framework, we successfully identify a new differential characteristic for the 2-round $\texttt{AES-128}$ $\texttt{fixed-TPKC}$ attack and develope the first practical $\texttt{fixed-TPKC}$ attack against 2-round $\texttt{AES-128}$. Additionally, we present practical $\texttt{fixed-TPKC}$ attacks against 5-round $\texttt{AES-192}$ and 3-round $\texttt{Kiasu-BC}$, along with a practical $\texttt{free-TPKC}$ attack against 6-round $\texttt{Kiasu-BC}$. Furthermore, we reduce time complexities for $\texttt{free-TPKC}$ and $\texttt{fixed-TPKC}$ attacks on other $\texttt{AES}$ variants.

Selfish mining attacks present a serious threat to Bitcoin security, enabling a miner with less than 51% of the network hashrate to gain higher rewards than their fair share. A growing body of works has studied the impact of such attacks and presented numerous strategies under a variety of model settings. This led to a complex landscape making it hard to comprehend the state of the art and distill insights, gaps, and trade-offs. In this paper, we demystify the landscape of selfish mining in Bitcoin by systematizing existing studies and evaluating more realistic model adaptations. To the best of our knowledge, our work is the first of its kind. We develop a multi-dimensional systematization framework assessing prior works based on their strategy formulation and targeted models. We go on to distill a number of insights and gaps highlighting open questions and understudied areas. Among them, we find that most of the surveyed works target the block-reward setting and do not account for transaction fees, and generally consider only single attackers. To bridge this gap, we evaluate several existing strategies in the transaction-fee regime---so miner's incentives come solely from transaction fees---for both single and multi-attacker scenarios. We also extend their models to include honest-but-rational miners showing how such adaptations could garner more performant strategy variations. Finally, we touch upon defenses proposed in the literature, and discuss connections between selfish mining and relevant incentivized/fee-driven paradigms.

Advanced Encryption Standard (AES) is one of the most widely used and extensively studied encryption algorithms globally, which is renowned for its efficiency and robust resistance to attacks. In this paper, three quantum circuits are designed to implement the S-box, which is the sole nonlinear component in AES. By incorporating a linear key schedule, we achieve a quantum circuit for implementing AES with the minimum number of qubits used. As a consequence, only 264/328/398 qubits are needed to implement the quantum circuits for AES-128/192/256. Furthermore, through quantum circuits of the S-box and key schedule, the overall size of the quantum circuit required for Grover's algorithm to attack AES is significantly decreased. This enhancement improves both the security and resource efficiency of AES in a quantum computing environment.

Recently, linear structures and algebraic attacks have been widely used in preimage attacks on round-reduced Keccak. Inherited by pioneers' work, we make some improvements for 3-round Keccak-256 and 4-round Keccak[r=640, c=160]. For 3-round Keccak-256, we introduce a three-stage model to deal with the unsatisfied restrictions while bringing more degrees of freedom at the same time. Besides, we show that guessing values for different variables will result in different complexity of solving time. With these techniques, the guessing times can be decreased to 2^{52}, and the solving time for each guess can be decreased to around 2^{5.2} 3-round Keccak calls. As a result, the complexity of finding a preimage for 3-round Keccak-256 can be decreased to around 2^{57.2}. For 4-round Keccak[r=640, c=160], an instance of the Crunchy Contest, we use some techniques to save degrees of freedom and make better linearization. Based on these techniques, we build an MILP model and obtain an attack with better complexity of around 2^{60.9}. The results of 3-round Keccak-256 and 4-round Keccak[r=640, c=160] are verified with real examples.

In this paper, we study preimage resistance of the SHA-3 standard. We propose a squeeze meet-in-the-middle attack as a new preimage attack method for the sponge functions. This attack combines the squeeze attack and meet-in-the-middle attack, and is implemented by internal differentials. We analyze the inverse operation of the SHA-3 round function, and develop a new target internal differential algorithm as well as a linearization technique for the Sbox in the backward phase. In addition, we propose the concept of a value-difference distribution table (VDDT) to optimize the attack complexity. These techniques lead to faster preimage attacks on five (out of six) SHA-3 functions reduced to 4 rounds, and also bring preimage attacks on 5 rounds of four SHA-3 instances. The attack techniques are verified by performing practical preimage attack on a small variant of 4-round Keccak.

Modern micro-architectural attacks use a variety of building blocks chained to develop a final exploit. However, since in most cases, the footprint of such attacks is not visible architecturally (like, in the file-system), it becomes trickier to defend against these. In light of this, several automated defence mechanisms use Hardware Performance Counters (HPCs) detect when the micro-architectural elements are being misused for a potential attacks (like flush-reload, Spectre, Meltdown etc.). In order to bypass such defences, recent works have proposed the idea of "probabilistic interleaving": the adversary interleaves the actual attack code with benign code with very low frequency. Such a strategy tips off the HPCs used for detection with a lot of unnecessary noise; recent studies have shown that probabilistically interleaved attacks can achieve an attack evasion rate of 100% (i.e. are virtually undetectable). In this work, we contend this folklore. We develop a theoretical model of interleaved attacks using lightweight statistical tools like Gaussian Mixture Models and Dip Test for Unimodality and prove they are detectable for the correct choices of HPCs. Furthermore, we also show possible defence strategy against a stronger threat model than considered in literature: where the attacker interleaves multiple attacks instead of a single attack. Empirically, to instantiate our detector, in contrast to prior detection strategies, we choose LLMs for a number of reasons: (1) LLMs can easily contextualize data from a larger set of HPCs than generic machine learning techniques, and (2) with simple prompts, LLMs can quickly switch between different statistical analysis methods. To this end, we develop an LLM-based methodology to detect probabilistically interleaved attacks. Our experiments establish that our improved methodology is able to achieve 100% speculative attacks like Spectre v1/v2/v3, Meltdown, and Spectre v2 (with improved gadgets that even evade recent protections like Enhanced IBRS, IBPB conditional, and so on). This makes our methodology suitable for detecting speculative attacks in a non-profiled setting: where attack signatures might not be known in advance. All in all, we achieve a 100% attack detection rate, even with very low interleave frequencies (i.e. $10^{-6}$). Our detection principle and its instantiation through LLMs shows how probabilistically interleaving attack code in benign execution is not a perfect strategy, and more research is still needed into developing and countering better attack evasion strategies.

In this paper, we analyze the subtle issues of complexity estimates related to state-of-the-art cryptanalytic efforts on ChaCha. In this regard, we demonstrate that the currently best-known cryptanalytic result on $7$-round ChaCha with time $2^{189.7}$ and data $2^{102.63}$ [Xu et al., ToSC 2024] can be estimated as $2^{178.12}$ for time and $2^{101.09}$ for data complexity. We improve the best-known result for the $7.25$ round by obtaining an improved set of Probabilistic Neutral Bits and considering our revised estimation. Our result with time complexity $2^{212.43}$ and data complexity $2^{100.56}$ improves the result of Xu et al., where they could achieve time and data complexity $2^{223.9}$ and $2^{100.80}$, respectively. For both the $7$ and $7.25$ rounds, we can show an improvement of the order of $2^{11}$ in the time complexity. For $7.5$-round, we improve the result of Dey [IEEE-IT 2024], which reports the time and data complexity of $2^{255.24}$ and $2^{32.64}$, respectively. By applying the formula of the same paper and incorporating additional PNBs, we obtain improved time and data complexity of $2^{253.23}$ and $2^{34.47}$, respectively. Thus, this paper describes the currently best-known cryptanalytic results against reduced round ChaCha. Our results do not affect the security claims of the complete algorithm with 20 rounds. Also, we provide a rebuttal of the Work by Wang et al. \cite{wangeprint} and analyze their claim about the error in the ``Divide-and-Conquer'' Approach.

This note demonstrates that the blind signature scheme based on cryptographic group actions, as proposed in ePrint paper 2025/397, fails to ensure blindness. Specifically, we construct an adversary that achieves a $1/8$ advantage in the blindness experiment. The attack leverages selective abort techniques (also known as selective failure attacks), a well-known strategy in the MPC literature.

We find an efficient method to solve the semidirect discrete logarithm problem (SDLP) over finite nonabelian groups of order $p^3$ and exponent $p^2$ for certain exponentially large parameters. This implies an attack on SPDH-Sign, a signature scheme based on the SDLP, for such parameters. We also take a step toward proving the quantum polynomial time equivalence of SDLP and SCDH.

The Shortest Vector Problem (SVP) is a cornerstone of lattice-based cryptography, underpinning the security of numerous cryptographic schemes like NTRU. Given its NP-hardness, efficient solutions to SVP have profound implications for both cryptography and computational complexity theory. This paper presents an innovative framework that integrates concepts from quantum gravity, noncommutative geometry, spectral theory, and post-supersymmetry (post-SUSY) particle physics to address SVP. By mapping high-dimensional lattice points to spinfoam networks and by means of Hamiltonian engineering, it is theoretically possible to devise new algorithms that leverage the interactions topologically protected Majorana fermion particles have with the gravitational field through the spectral action principle to loop through these spinfoam networks where SVP vectors could then be encoded onto the spectrum of the corresponding Dirac-like dilation operators within the system. We establish a novel approach that leverages post-SUSY physics and theories of quantum gravity to achieve algorithmic speedups beyond those expected by conventional quantum computers. This interdisciplinary methodology not only proposes potential polynomial-time algorithms for SVP, but also bridges gaps between theoretical physics and cryptographic applications, providing further insights into the Riemann Hypothesis (RH) and the Hilbert-Pólya Conjecture. Possible directions for experimental realization through biologically inspired hardware or biological tissues by orchestrated objective reduction (Orch-Or) theory are discussed.

Proving resistance to conventional attacks, e.g., differential, linear, and integral attacks, is essential for designing a secure symmetric-key cipher. Recent advances in automatic search and deep learning-based methods have made this time-consuming task relatively easy, yet concerns persist over expertise requirements and potential oversights. To overcome these concerns, Kimura et al. proposed neural network-based output prediction (NN) attacks, offering simplicity, generality, and reduced coding mistakes. NN attacks could be helpful for designing secure symmetric-key ciphers, especially the S-box-based block ciphers. Inspired by their work, we first apply NN attacks to Simon, one of the AND-Rotation-XOR-based block ciphers, and identify structures susceptible to NN attacks and the vulnerabilities detected thereby. Next, we take a closer look at the vulnerable structures. The most vulnerable structure has the lowest diffusion property compared to others. This fact implies that NN attacks may detect such a property. We then focus on a biased event of the core function in vulnerable Simon-like ciphers and build effective linear approximations caused by such an event. Finally, we use these linear approximations to reveal that the vulnerable structures are more susceptible to a linear key recovery attack than the original one. We conclude that our analysis can be a solid step toward making NN attacks a helpful tool for designing a secure symmetric-key cipher.

Intel Trust Domain Extensions (TDX) has emerged as a crucial technology aimed at strengthening the isolation and security guarantees of virtual machines, especially as the demand for secure computation is growing largely. Despite the protections offered by TDX, in this work, we dig deep into the security claims and uncover an intricate observation in TDX. These findings undermine TDX's core security guarantees by breaching the isolation between the Virtual Machine Manager (VMM) and Trust Domains (TDs). In this work for the first time, we show through a series of experiments that these performance counters can also be exploited by the VMM to differentiate between activities of an idle and active TD. The root cause of this leakage is core contention. This occurs when the VMM itself, or a process executed by the VMM, runs on the same core as the TD. Due to resource contention on the core, the effects of the TD's computations become observable in the performance monitors collected by the VMM. This finding underscore the critical need for enhanced protections to bridge these gaps within these advanced virtualized environments.

We examine the problem of finding small solutions to systems of modular multivariate polynomials. While the case of univariate polynomials has been well understood since Coppersmith's original 1996 work, multivariate systems typically rely on carefully crafted shift polynomials and significant manual analysis of the resulting Coppersmith lattice. In this work, we develop several algorithms that make such hand-crafted strategies obsolete. We first use the theory of Gröbner bases to develop an algorithm that provably computes an optimal set of shift polynomials, and we use lattice theory to construct a lattice which provably contains all desired short vectors. While this strategy is usable in practice, the resulting lattice often has large rank. Next, we propose a heuristic strategy based on graph optimization algorithms that quickly identifies low-rank alternatives. Third, we develop a strategy which symbolically precomputes shift polynomials, and we use the theory of polytopes to polynomially bound the running time. Like Meers and Nowakowski's automated method, our precomputation strategy enables heuristically and automatically determining asymptotic bounds. We evaluate our new strategies on over a dozen previously studied Coppersmith problems. In all cases, our unified approach achieves the same recovery bounds in practice as prior work, even improving the practical bounds for three of the problems. In five problems, we find smaller and more efficient lattice constructions, and in three problems, we improve the existing asymptotic bounds. While our strategies are still heuristic, they are simple to describe, implement, and execute, and we hope that they drastically simplify the application of Coppersmith's method to systems of multivariate polynomials.

Given its integral role in modern encryption systems such as CRYSTALS-Kyber, the Fujisaki-Okamoto (FO) transform will soon be at the center of our secure communications infrastructure. An enduring debate surrounding the FO transform is whether to use explicit or implicit rejection when decapsulation fails. Presently, implicit rejection, as implemented in CRYSTALS-Kyber, is supported by a strong set of arguments. Therefore, understanding its security implications in different attacker models is essential. In this work, we study implicit rejection through a novel lens, namely, from the perspective of kleptography. Concretely, we consider an attacker model in which the attacker can subvert the user's code to compromise security while remaining undetectable. In this scenario, we present three attacks that significantly reduce the security level of the FO transform with implicit rejection.

Multi-Party Computation (MPC) has become a major tool for protecting hundreds of billions of dollars in cryptocurrency wallets. MPC protocols are currently powering the wallets of Coinbase, Binance, Zengo, BitGo, Fireblocks and many other fintech companies servicing thousands of financial institutions and hundreds of millions of end-user consumers. We present four novel key-extraction attacks on popular MPC signing protocols showing how a single corrupted party may extract the secret in full during the MPC signing process. Our attacks are highly practical (the practicality of the attack depends on the number of signature-generation ceremonies the attacker participates in before extracting the key). Namely, we show key-extraction attacks against different threshold-ECDSA protocols/implementations requiring $10^6$, $256$, $16$, and *one signature*, respectively. In addition, we provide proof-of-concept code that implements our attacks.

Secure two-party comparison, known as Yao's millionaires' problem, has been a fundamental challenge in privacy-preserving computation. It enables two parties to compare their inputs without revealing the exact values of those inputs or relying on any trusted third party. One elegant approach to secure computation is based on homomorphic encryption. Recently, building on this approach, Carlton et al. (CT-RSA 2018) and Bourse et al. (CT-RSA 2020) presented novel solutions for the problem of secure integer comparison. These protocols have demonstrated significantly improved performance compared to the well-known and frequently used DGK protocol (ACISP 2007 and Int. J. Appl. Cryptogr. 1(4),323–324, 2009). In this paper, we introduce a class of higher residuosity attacks, which can be regarded as an extension of the classical quadratic residuosity attack on the decisional Diffie-Hellman problem. We demonstrate that the small RSA subgroup decision problems, upon which both the CEK and BST protocols are based, are not difficult to solve when the prime base $p_0$ is small (e.g., \( p_0 < 100 \)). Under these conditions, the protocols achieve optimal overall performance. Furthermore, we offer recommendations for precluding such attacks, including one approach that does not adversely affect performance. We hope that these attacks can be applied to analyze other number-theoretic hardness assumptions.

We study the hardness of the Syndrome Decoding problem, the base of most code-based cryptographic schemes, such as Classic McEliece, in the presence of side-channel information. We use ChipWhisperer equipment to perform a template attack on Classic McEliece running on an ARM Cortex-M4, and accurately classify the Hamming weights of consecutive 32-bit blocks of the secret error vector. With these weights at hand, we optimize Information Set Decoding algorithms. Technically, we show how to speed up information set decoding via a dimension reduction, additional parity-check equations, and an improved information set search, all derived from the Hamming weight information. Consequently, using our template attack, we can practically recover an error vector in dimension n=2197 in a matter of seconds. Without side-channel information, such an instance has a complexity of around 88 bit. We also estimate how our template attack affects the security of the proposed McEliece parameter sets. Roughly speaking, even an error-prone leak of our Hamming weight information leads for n=3488 to a security drop of 89 bits.

We point out flaw in zero-knowledge of the CROSS identification protocol, $\textsf{CROSS-ID}$, which allows a distinguisher to distinguish real and simulated transcripts given access to the witness. Moreover, we show that the real and simulated transcripts are not statistically indistinguishable, and therefore the protocol can only satisfy weak computational (rather than strong, statistical or perfect) Honest Verifier Zero-knowledge. This issue is still present in version 2.0 updated on January 31, 2025, which resolves the security losses attained via the attacks of [BLP+25]

Falcon is one of post-quantum signature schemes selected by NIST for standardization. With the deployment underway, its implementation security is of great importance. In this work, we focus on the side-channel security of Falcon and our contributions are threefold. First, by exploiting the symplecticity of NTRU and a recent decoding technique, we dramatically improve the key recovery using power leakages within Falcon Gaussian samplers. Compared to the state of the art (Zhang, Lin, Yu and Wang, EUROCRYPT 2023), the amount of traces required by our attack for a full key recovery is reduced by at least 85%. Secondly, we present a complete power analysis for two exposed power leakages within Falcon’s integer Gaussian sampler. We identify new sources of these leakages, which have not been identified by previous works, and conduct detailed security evaluations within the reference implementation of Falcon on Chipwhisperer. Thirdly, we propose effective and easy-to-implement countermeasures against both two leakages to protect the whole Falcon’s integer Gaussian sampler. Configured with our countermeasures, we provide security evaluations on Chipwhisperer and report performance of protected implementation. Experimental results highlight that our countermeasures admit a practical trade-off between effciency and side-channel security.

SNOVA is a multivariate signature scheme submitted to the NIST project for additional signature schemes by Cho, Ding, Kuan, Li, Tseng, Tseng, and Wang. With small key and signature sizes good performance, SNOVA is one of the more efficient schemes in the competition, which makes SNOVA an important target for cryptanalysis. In this paper, we observe that SNOVA implicitly uses a structured version of the ``whipping'' technique developed for the MAYO signature scheme. We show that the extra structure makes the construction vulnerable to new forgery attacks. Concretely, we formulate new attacks that reduce the security margin of the proposed SNOVA parameter sets by a factor between $2^{8}$ and $2^{39}$. Furthermore, we show that large fractions of public keys are vulnerable to more efficient versions of our attack. For example, for SNOVA-37-17-2, a parameter set targeting NIST's first security level, we show that roughly one out of every $500$ public keys is vulnerable to a universal forgery attack with bit complexity $2^{97}$, and roughly one out of every $143000$ public keys is even breakable in practice within a few minutes.

This paper presents a novel single-trace side-channel attack on FALCON—a lattice-based post-quantum digital signature protocol recently approved for standardization by NIST. We target the discrete Gaussian sampling operation within the FALCON key generation scheme and use a single power measurement trace to succeed. Notably, negating the ‘shift right 63-bit’ operation (for 64-bit values) leaks critical information about the ‘-1’ vs. ‘0’ assignments to intermediate coefficients. These leaks enable full recovery of the generated secret keys. The proposed attack is simulated on the ELMO simulator running both reference and optimized software implementation from FALCON’s NIST Round 3 package. Statistical analysis with 20k tests reveals a full key-recovery success rate of 100% for FALCON-512. This work highlights the vulnerability of current software solutions to single-trace attacks and underscores the urgent need to develop single-trace resilient software for embedded systems in the presilicon phase.

We revisit the polynomial attack to the $\mathsf{ROS}$ problem modulo $p$ from [BLLOR22]. Our new algorithm achieves a polynomial time solution in dimension $\ell \gtrsim 0.725 \cdot \log_2 p$, extending the range of dimensions for which a polynomial attack is known beyond the previous bound of $\ell > \log_2p$. We also combine our new algorithm with Wagner's attack to improve the general $\mathsf{ROS}$ attack complexity for some of the dimensions where a polynomial solution is still not known. We implement our polynomial attack and break the one-more unforgeability of blind Schnorr signatures over 256-bit elliptic curves in a few seconds with 192 concurrent sessions.

The Gentry-Peikert-Vaikuntanathan (GPV) framework is utilized for constructing practical digital signatures, which is proven to be secure in both the classical/quantum random-oracle models. Falcon is such a signature scheme, recognized as a compact and efficient signature among NIST-standardized signature schemes. Recently, Guerreau et al. (CHES 2022) and Zhang et al. (Eurocrypt 2023) proposed the secret key recovery attack on Falcon utilizing signatures filtered by simple power analysis (SPA) attacks. However, these attacks, which exploit the conditional signature distributions, require a large number of SPA attacks to obtain the filtered signatures. Furthermore, no existing attack considers general GPV signatures despite the importance of the GPV framework in modern digital signatures. Therefore, we address these problems as follows. First, we introduce, for the first time, a concept of vulnerable partial information of GPV signatures and propose a non-filtering secret key recovery attack, called OLS attack, which effectively utilizes partial information without filtering. The proposed OLS attack is a linear estimator with computational complexity that scales linearly with the number of samples, making OLS attack highly practical. Furthermore, we prove that the secret key recovered by the OLS attack converges to the real secret key in probability as the number of samples increases. Second, we leverage SPA to extract Gaussian leakage, which is used as partial information for the OLS attack on Falcon. As a result, the OLS attack shows a significantly higher success rate with the fewest samples than the state-of-the-art attacks. Furthermore, by incorporating the DDGR attack, the OLS attack can recover the secret key using much fewer samples with a success rate close to 100%. Moreover, we propose an OLS attack specialized for Falcon, which can even more reduce the number of required side-channel attacks. Third, we propose an error-tolerant power analysis attack using MAP decoding, which effectively corrects the errors in samples to properly estimate Gaussian leakage. For concrete experimental validation, an ELMO simulator generates noisy power traces and ChipWhisperer measures power traces from the STM32F415 model. The proposed MAP decoding achieves high effectiveness for estimating Gaussian leakage, particularly when applied to power traces collected using low-resolution ChipWhisperer. In conclusion, it is important for future work to study countermeasures for OLS attacks.

Over the past decade and a half, cryptanalytic techniques for Salsa20 have been increasingly refined, largely following the overarching concept of Probabilistically Neutral Bits (PNBs) by Aumasson et al. (FSE 2008). In this paper, we present a novel criterion for choosing key-$\mathcal{IV}$ pairs using certain 2-round criteria and connect that with clever tweaks of existing techniques related to Probabilistically Independent $\mathcal{IV}$ bits (earlier used for ARX ciphers, but not for Salsa20) and well-studied PNBs. Through a detailed examination of the matrix after initial rounds of Salsa20, we introduce the first-ever cryptanalysis of Salsa20 exceeding $8$ rounds. Specifically, Salsa20/$8.5$, consisting of $256$ secret key bits, can be cryptanalyzed with a time complexity of $2^{245.84}$ and data amounting to $2^{99.47}$. Further, the sharpness of our attack can be highlighted by showing that Salsa20/$8$ can be broken with time $2^{186.01}$ and data $2^{99.73}$, which is a significant improvement over the best-known result of Coutinho et al. (Journal of Cryptology, 2023, time $2^{217.14}$ and data $2^{113.14}$). Here, the refinements related to backward biases for PNBs are also instrumental in achieving the improvements. We also provide certain instances of how these ideas improve the cryptanalysis on $128$-bit versions. In the process, a few critical points are raised on some existing state-of-the-art works in this direction, and in those cases, their estimates of time and data are revisited to note the correct complexities, revising the incorrect numbers.

In this article we present a non-uniform reduction from rank- 2 module-LIP over Complex Multiplication fields, to a variant of the Principal Ideal Problem, in some fitting quaternion algebra. This reduction is classical deterministic polynomial-time in the size of the inputs. The quaternion algebra in which we need to solve the variant of the principal ideal problem depends on the parameters of the module-LIP problem, but not on the problem’s instance. Our reduction requires the knowledge of some special elements of this quaternion algebras, which is why it is non-uniform. In some particular cases, these elements can be computed in polynomial time, making the reduction uniform. This is the case for the Hawk signature scheme: we show that breaking Hawk is no harder than solving a variant of the principal ideal problem in a fixed quaternion algebra (and this reduction is uniform).

The rank-$2$ module-LIP problem was introduced in cryptography by (Ducas, Postlethwaite, Pulles, van Woerden, Asiacrypt 2022), to construct the highly performant HAWK scheme. A first cryptanalytic work by (Mureau, Pellet--Mary, Pliatsok, Wallet, Eurocrypt 2024) showed a heuristic polynomial time attack against the rank-$2$ module-LIP problem over totally real number fields. While mathematically interesting, this attack focuses on number fields that are not relevant for cryptography. The main families of fields used in cryptography are the highly predominant cyclotomic fields (used for instance in the HAWK scheme), as well as the NTRU Prime fields, used for instance in the eponymous NTRU Prime scheme (Bernstein, Chuengsatiansup, Lange, van Vredendaal, SAC 2017). In this work, we generalize the attack of Mureau et al. against rank-$2$ module-LIP to the family of all number fields with at least one real embedding, which contains the NTRU Prime fields. We present three variants of our attack, firstly a heuristic one that runs in quantum polynomial time. Secondly, under the extra assumption that the defining polynomial of $K$ has a $2$-transitive Galois group (which is the case for the NTRU Prime fields), we give a provable attack that runs in quantum polynomial time. And thirdly, with the same $2$-transitivity assumption we give a heuristic attack that runs in classical polynomial time. For the latter we use a generalization of the Gentry--Szydlo algorithm to any number field which might be of independent interest.

The study of attack algorithms for the Learning with Errors (LWE) problem is crucial for the cryptanalysis of LWE-based cryptosystems. The BKW algorithm has gained significant attention as an important combinatorial attack for solving LWE. However, its exponential time and memory requirements severely limit its practical applications, even with medium-sized parameters. In this paper, we present a memory-efficient BKW algorithm for LWE, which extends Bogos's work [Asiacrypt'16] on the Learning Parity with Noise (LPN) problem. While their work improved efficiency, it overlooked the high memory demands of the BKW algorithm. We address this with two key improvements. First, we propose an efficient reduction technique for low-memory regimes, \(c\)-sum-PCS-reduce, which combines the \(c\)-sum technique with Parallel Collision Search (PCS) to achieve a better time-memory trade-off. Second, we present an improved memory-optimized finite automaton for our optimized BKW algorithm by incorporating several efficient memory-saving reduction techniques and pruning potential high-memory paths. Our algorithm, using graphs as a meta tool, can automatically identify the optimal reduction path within the graph, aiming to reduce both time and memory complexities. Compared to the state-of-the-art coded-BKW in the lattice-estimator, our algorithm achieves time complexity improvements ranging from \(2^{3.3}\) to \(2^{26.2}\). Furthermore, memory complexity is improved, with reductions ranging from \(2^{9.7}\) to \(2^{71.3}\).

Lattice-based cryptography is in the process of being standardized. Several proposals to deal with side-channel information using lattice reduction exist. However, it has been shown that algorithms based on Bayesian updating are often more favorable in practice. In this work, we define distribution hints; a type of hint that allows modelling probabilistic information. These hints generalize most previously defined hints and the information obtained in several attacks. We define two solvers for our hints; one is based on belief propagation and the other one uses a greedy approach. We prove that the latter is a computationally less expensive approximation of the former and that previous algorithms used for specific attacks may be seen as special cases of our solvers. Thereby, we provide a systematization of previously obtained information and used algorithms in real-world side-channel attacks. In contrast to lattice-based approaches, our framework is not limited to value leakage. For example, it can deal with noisy Hamming weight leakage or partially incorrect information. Moreover, it improves upon the recovery of the secret key from approximate hints in the form they arise in real-world attacks. Our framework has several practical applications: We exemplarily show that a recent attack can be improved; we reduce the number of traces and corresponding ciphertexts and increase the noise resistance. Further, we explain how distribution hints could be applied in the context of previous attacks and outline a potential new attack.

SNOVA is a post-quantum digital signature scheme based on multivariate polynomials. It is a second-round candidate in an ongoing NIST standardization process for post-quantum signatures, where it stands out for its efficiency and compactness. Since its initial submission, there have been several improvements to its security analysis, both on key recovery and forgery attacks. All these works reduce to solving a structured system of quadratic polynomials, which we refer to as SNOVA system. In this work, we propose a polynomial solving algorithm tailored for SNOVA systems, which exploits the \textit{stability} of the system under the action of a commutative group of matrices. This new algorithm reduces the complexity of solving SNOVA systems over generic ones. We show how to adapt the \textit{reconciliation} and \textit{direct} attacks in order to profit from the new algorithm. Consequently, we improve the reconciliation attack for all SNOVA parameter sets with speedup factors ranging between $2$ and $2^{20}$. We also show how to use similar ideas to carry on a forgery attack. In this case, we use experimental results to estimate its complexity, and we discuss its impact. The empirical evidence suggests that our attack is more efficient than previous attacks, and it takes some SNOVA parameter sets below NIST's security threshold.

In this work, we study the singular locus of the varieties defined by the public keys of UOV and VOX, two multivariate signature schemes submitted to the additional NIST call for post-quantum signature schemes. We give a new attack for $\hat +$ and VOX targeting singular points of the underlying UOV key. Our attack lowers the security of the schemes, both asymptotically and in number of gates, showing in particular that the parameter sets proposed for these schemes do not meet the NIST security requirements. More precisely, we show that the security of VOX/UOV$\hat +$ was overestimated by factors $2^{2}, 2^{18}, 2^{37}$ for security levels I, III, V respectively. As an essential element of the attack on VOX, we introduce a polynomial time algorithm performing a key recovery from one vector, with an implementation requiring only $15$ seconds at security level V.

SNOVA, a post-quantum signature scheme with compact key sizes, is a second-round NIST candidate. This paper conducts a fault analysis of SNOVA, targeting permanent and transient faults during signature generation. We propose fault injection strategies that exploit SNOVA's structure, enabling key recovery with as few as $22$ to $68$ faulty signatures, depending on security levels. A novel fault-assisted reconciliation attack is introduced that effectively extracts the secret key space by solving a quadratic polynomial system. Simulations reveal that transient or permanent faults in signature generation can severely compromise security. We also suggest a lightweight countermeasure to mitigate fault attacks with minimal overhead. Our findings emphasize the need for fault-resistant mechanisms in post-quantum schemes like SNOVA.

ML-KEM and ML-DSA are NIST-standardized lattice-based post-quantum cryptographic algorithms. In both algorithms, Keccak is the designated hash algorithm extensively used for deriving sensitive information, making it a valuable target for attackers. In the field of fault injection attacks, few works targeted Keccak, and they have not fully explored its impact on the security of ML-KEM and ML-DSA. Consequently, many attacks remain undiscovered. In this article, we first identify various fault vulnerabilities of KECCAK that determine the (partial) output by manipulating the control flow under a practical loop-abort model. Then, we systematically analyze the impact of a faulty Keccak output and propose six attacks against ML-KEM and five attacks against ML-DSA, including key recovery, signature forgery, and verification bypass. These attacks cover the key generation, encapsulation, decapsulation, signing, and verification phases, making our scheme the first to apply to all phases of ML-KEM and ML-DSA. The proposed attacks are validated on the C implementations of the PQClean library’s ML-KEM and ML-DSA running on embedded devices. Experiments show that the required loop-abort faults can be realized on ARM Cortex-M0+, M3, M4, and M33 microprocessors with low-cost electromagnetic fault injection settings, achieving a success rate of 89.5%. Once the fault injection is successful, all proposed attacks can succeed with a probability of 100%.

The McEliece cryptosystem is a strong contender for post-quantum schemes, including key encapsulation for confidentiality of key exchanges in network protocols. A McEliece secret key is a structured parity check matrix that is transformed via Gaussian elimination into an unstructured public key. We show that this transformation is highly critical with respect to side-channel leakage. We assume leakage of the elementary row operations during Gaussian elimination, motivated by McEliece implementations in the cryptographic libraries Classic McEliece and Botan. We propose a novel decoding algorithm to reconstruct a secret key from its public key with information from a Gaussian transformation leak. Even if the obtained side-channel leakage is extremely noisy, i.e., each bit is flipped with probability as high as τ ≈ 0.4, we succeed to recover the secret key in a matter of minutes for all proposed (Classic) McEliece instantiations. Remarkably, for high-security McEliece parameters, our attack is more powerful in the sense that it can tolerate even larger τ. We demonstrate our attack on the constant-time reference implementation of Classic McEliece in a single-trace setting, using an STM32L592 ARM processor. Our result stresses the necessity of properly protecting highly structured code-based schemes such as McEliece against side-channel leakage.

In Dilithium, the rejection sampling step is crucial for the proof of security and correctness of the scheme. However, to our knowledge, there is no attack in the literature that takes advantage of an attacker knowing rejected signatures. The aim of this paper is to create a practical black-box attack against Dilithium with a weakened rejection sampling. We succeed in showing that an adversary with enough rejected signatures can recover Dilithium's secret key in less than half an hour on a desktop computer. There is one possible application for this result: by physically preventing one of the rejection sampling tests from happening, we obtain two fault attacks against Dilithium.

We present several new provable algorithms for two variants of the code equivalence problem on linear error-correcting codes, the Linear Code Equivalence Problem (LCE) and the Permutation Code Equivalence Problem (PCE). Specifically, for arbitrary codes of block length $n$ and dimension $k$ over any finite field $\mathbb{F}_q$, we show: 1) A deterministic algorithm running in $2^{n + o(n+q)}$ time for LCE. 2) A randomized algorithm running in $2^{n/2 + o(n+q)}$ time for LCE and PCE. 3) A quantum algorithm running in $2^{n/3 + o(n+q)}$ time for LCE and PCE. The first algorithm complements the deterministic roughly $2^n$-time algorithm of Babai (SODA 2011) for PCE. The second two algorithms improve on recent work of Nowakowski (PQCrypto 2025), which gave algorithms with similar running times, but only for code equivalence on \emph{random} codes and only over fields of order $q \geq 7$.

Frontrunning is rampant in blockchain ecosystems, yielding attackers profits that have already soared into several million. Most existing frontrunning attacks focus on manipulating transaction order (namely, prioritizing attackers' transactions before victims' transactions) $\textit{within}$ a block. However, for the emerging directed acyclic graph (DAG)-based blockchains, these intra-block frontrunning attacks may not fully reveal the frontrunning vulnerabilities as they introduce block ordering rules to order transactions belonging to distinct blocks. This work performs the first in-depth analysis of frontrunning attacks toward DAG-based blockchains. We observe that the current block ordering rule is vulnerable to a novel $\textit{inter-block}$ frontrunning attack, which enables the attacker to prioritize ordering its transactions before the victim transactions across blocks. We introduce three attacking strategies: $\textit{(i)}$ Fissure attack, where attackers render the victim transactions ordered later by disconnecting the victim's blocks. $\textit{(ii)}$ Speculative attack, where attackers speculatively construct order-priority blocks. $\textit{(iii)}$ Sluggish attack, where attackers deliberately create low-round blocks assigned a higher ordering priority by the block ordering rule. We implement our attacks on two open-source DAG-based blockchains, Bullshark and Tusk. We extensively evaluate our attacks in geo-distributed AWS and local environments by running up to $n=100$ nodes. Our experiments show remarkable attack effectiveness. For instance, with the speculative attack, the attackers can achieve a $92.86\%$ attack success rate (ASR) on Bullshark and an $86.27\%$ ASR on Tusk. Using the fissure attack, the attackers can achieve a $94.81\%$ ASR on Bullshark and an $87.31\%$ ASR on Tusk. We also discuss potential countermeasures for the proposed attack, such as ordering blocks randomly and reordering transactions globally based on transaction fees. However, we find that they either compromise the performance of the system or make the protocol more vulnerable to frontrunning using the existing frontrunning strategies.

Multivariate cryptography currently centres mostly around UOV-based signature schemes: All multivariate round 2 candidates in the selection process for additional digital signatures by NIST are either UOV itself or close variations of it: MAYO, QR-UOV, SNOVA, and UOV. Also schemes which have been in the focus of the multivariate research community, but are broken by now - like Rainbow and LUOV - are based on UOV. Both UOV and the schemes based on it have been frequently analyzed regarding their physical security in the course of the NIST process. However, a comprehensive analysis regarding the physical security of UOV-based signature schemes is missing. In this work, we want to bridge this gap and create a comprehensive overview of physical attacks on UOV and its variants from the second round of NIST’s selection process for additional post-quantum signature schemes, which just started. First, we collect all existing side-channel and fault attacks on UOV-based schemes and transfer them to the current UOV specification. Since UOV was subject to significant changes over the past few years, e.g., adaptions to the expanded secret key, some attacks need to be reassessed. Next, we introduce new physical attacks in order to obtain an overview as complete as possible. We then show how all these attacks would translate to MAYO, QR-UOV, and SNOVA. To improve the resistance of UOV-based signature schemes towards physical attacks, we discuss and introduce dedicated countermeasures. As related result, we observe that certain implementation decisions, like key compression techniques and randomization choices, also have a large impact on the physical security, in particular on the effectiveness of the considered fault attacks. Finally, we provide implementations of UOV and MAYO for the ARM Cortex-M4 architecture that feature first-order masking and protection against selected fault attacks. We benchmark the resulting overhead on a NUCLEO-L4R5ZI board and validate our approach by performing a TVLA on original and protected subroutines, yielding significantly smaller t-values for the latter.

Recent server-side optimizations like speculative decoding significantly enhance the interactivity and resource efficiency of Large Language Model (LLM) services. However, we show that these optimizations inadvertently introduce new side-channel vulnerabilities through network packet timing and size variations that tend to be input-dependent. Network adversaries can leverage these side channels to learn sensitive information contained in \emph{encrypted} user prompts to and responses from public LLM services. This paper formalizes the security implications using a novel indistinguishability framework and introduces a novel attack that establishes the insecurity of real-world LLM services with streaming APIs under our security framework. Our proposed attack effectively deconstructs encrypted network packet traces to reveal the sizes of underlying LLM-generated tokens and whether the tokens were generated with or without certain server-side optimizations. Our attack can accurately predict private attributes in real-world privacy-sensitive LLM applications in medicine and finance with $71$--$92\%$ accuracy on an open-source vLLM service and $50$--$90\%$ accuracy on the commercial ChatGPT service. Finally, we show that solutions that hide these side channels to different degrees expose a tradeoff between security and performance --- specifically, interactivity and network bandwidth overheads.

Electromagnetic side-channel analysis is a powerful method for monitoring processor activity and compromising cryptographic systems in air-gapped environments. As analytical methodologies and target devices evolve, the importance of leakage localization and probe aiming becomes increasingly apparent for capturing only the desired signals with a high signal-to-noise ratio. Despite its importance, there remains substantial reliance on unreliable heuristic approaches and inefficient exhaustive searches. Furthermore, related studies often fall short in terms of feasibility, practicality, and performance, and are limited to controlled DUTs and low-end MCUs. To address the limitations and inefficiencies of the previous approaches, we propose a novel methodology―${\rm P{\tiny ROBE}S{\tiny HOOTER}}$―for leakage localization and probe aiming. This approach leverages new insights into the spatial characteristics of amplitude modulation and intermodulation distortion in processors. As a result, ${\rm P{\tiny ROBE}S{\tiny HOOTER}}$ provides substantial improvements in various aspects: $\boldsymbol 1)$ it is applicable to not only simple MCUs but also complex SoCs, $\boldsymbol 2)$ it effectively handles multi-core systems and dynamic frequency scaling, $\boldsymbol 3)$ it is adoptable to uncontrollable DUTs, making it viable for constrained real-world attacks, and $\boldsymbol 4)$ it performs significantly faster than previous methods. To demonstrate this, we experimentally evaluate ${\rm P{\tiny ROBE}S{\tiny HOOTER}}$ on a high-end MCU (the NXP i.MX RT1061 featuring a single ARM Cortex-M7 core) and a complex SoC (the Broadcom BCM2711 equipped with the Raspberry Pi 4 Model B, featuring four ARM Cortex-A72 cores).

We study the linear code equivalence problem (LEP) for linear $[n,k]$-codes over finite fields $\mathbb{F}_q$. Recently, Chou, Persichetti and Santini gave an elegant algorithm that solves LEP over large finite fields (with $q = \Omega(n)$) in time $2^{\frac{1}{2}\operatorname{H}\left(\frac{k}{n}\right)n}$, where $\operatorname{H}(\cdot)$ denotes the binary entropy function. However, for small finite fields, their algorithm can be significantly slower. In particular, for fields of constant size $q = \mathcal{O}(1)$, its runtime increases by an exponential factor in $n$. We present an improved version of their algorithm, which achieves the desired runtime of $2^{\frac{1}{2}\operatorname{H}\left(\frac{k}{n}\right)n}$ for all finite fields of size $q \geq 7$. For a wide range of parameters, this improves over the runtime of all previously known algorithms by an exponential factor.

This paper presents enhanced reductions of the bounded-weight and exact-weight Syndrome Decoding Problem (SDP) to a system of quadratic equations. Over $\mathbb{F}_2$, we improve on a previous work and study the degree of regularity of the modeling of the exact weight SDP. Additionally, we introduce a novel technique that transforms SDP instances over $\mathbb{F}_q$ into systems of polynomial equations and thoroughly investigate the dimension of their varieties. Experimental results are provided to evaluate the complexity of solving SDP instances using our models through Gröbner bases techniques.

We present a key-recovery attack on DEFI, an efficient signature scheme proposed recently by Feussner and Semaev, and based on isotropic quadratic forms, borrowing from both multivariate and lattice cryptography. Our lattice-based attack is partially heuristic, but works on all proposed parameters: experimentally, it recovers the secret key in a few minutes, using less than ten (message,signature) pairs.

Lattice-based cryptography typically uses lattices with special properties to improve efficiency. We show how blockwise reduction can exploit lattices with special geometric properties, effectively reducing the required blocksize to solve the shortest vector problem to half of the lattice's rank, and in the case of the hypercubic lattice $\mathbb{Z}^n$, further relaxing the approximation factor of blocks to $\sqrt{2}$. We study both provable algorithms and the heuristic well-known primal attack, in the case where the lattice has a first minimum that is almost as short as that of the hypercubic lattice $\mathbb{Z}^n$. Remarkably, these near-hypercubic lattices cover Falcon and most concrete instances of the NTRU cryptosystem: this is the first provable result showing that breaking NTRU lattices can be reduced to finding shortest lattice vectors in halved dimension, thereby providing a positive response to a conjecture of Gama, Howgrave-Graham and Nguyen at Eurocrypt 2006. Yet, the best primal attack on NTRU heuristically decreases the $1/2$ provable dimension reduction factor to $4/9$.

The Performance Monitoring Unit (PMU), a standard feature in all modern computing systems, presents significant security risks by leaking sensitive user activities through microarchitectural event data. This work demonstrates the feasibility of remote side-channel attacks leveraging PMU data, revealing vulnerabilities that compromise user privacy and enable covert surveillance without physical access to the target machine. By analyzing the PMU feature space, we create distinct micro-architectural fingerprints for benchmark applications, which are then utilized in machine learning (ML) models to detect the corresponding benchmarks. This approach allows us to build a pre-trained model for benchmark detection using the unique micro-architectural fingerprints derived from PMU data. Subsequently, when an attacker remotely accesses the victim’s PMU data, the pre-trained model enables the identification of applications used by the victim with high accuracy. In our proof-of-concept demonstration, the pre-trained model successfully identifies applications used by a victim when the attacker remotely accesses PMU data, showcasing the potential for malicious exploitation of PMU data. We analyze stress-ng benchmarks and build our classifiers using logistic regression, decision tree, k-nearest neighbors, and random forest ML models. Our proposed models achieve an average prediction accuracy of 98%, underscoring the potential risks associated with remote side-channel analysis using PMU data and emphasizing the need for more robust safeguards. This work underscores the urgent need for robust countermeasures to protect against such vulnerabilities and provides a foundation for future research in micro-architectural security.

Twisted generalized Reed-Solomon (TGRS) codes constitute an interesting family of evaluation codes, containing a large class of maximum distance separable codes non-equivalent to generalized Reed-Solomon (GRS) ones. Moreover, the Schur squares of TGRS codes may be much larger than those of GRS codes with same dimension. Exploiting these structural differences, in 2018, Beelen, Bossert, Puchinger and Rosenkilde proposed a subfamily of Maximum Distance Separable (MDS) Twisted Reed--Solomon (TRS) codes over $\mathbb{F}_q$ with $\ell$ twists $q \approx n^{2^{\ell}}$ for McEliece encryption, claiming their resistance to both Sidelnikov Shestakov attack and Schur products--based attacks. In short, they claimed these codes to resist to classical key recovery attacks on McEliece encryption scheme instantiated with Reed-Solomon (RS) or GRS codes. In 2020, Lavauzelle and Renner presented an original attack on this system based on the computation of the subfield subcode of the public TRS code. In this paper, we show that the original claim on the resistance of TRS and TGRS codes to Schur products based--attacks is wrong. We identify a broad class of codes including TRS and TGRS ones that is distinguishable from random by computing the Schur square of some shortening of the code. Then, we focus on the case of single twist ({i.e.}, $\ell = 1$), which is the most efficient one in terms of decryption complexity, to derive an attack. The technique is similar to the distinguisher-based attacks of RS code-based systems given by Couvreur, Gaborit, Gauthier-Umaña, Otmani, Tillich in 2014.

Unbalanced Oil and Vinegar (UOV) is one of the oldest, simplest, and most studied ad-hoc multivariate signature schemes. UOV signature schemes are attractive because they have very small signatures and fast verification. On the downside, they have large public and secret keys. As a result, variations of the traditional UOV scheme are usually developed with the goal to reduce the key sizes. Seven variants of UOV were submitted to the additional call for digital signatures by NIST, prior to which, a variant named MQ-Sign was submitted to the (South) Korean post-quantum cryptography competition (KpqC). MQ-Sign is currently competing in the second round of KpqC with two variants. One of the variants corresponds to the classic description of UOV with certain implementation and parameter choices. In the other variant, called MQ-Sign-LR, a part of the central map is constructed from row shifts of a single matrix. This design makes for smaller secret keys, and in the case where the equivalent keys optimization is used, it also leads to smaller public keys. However, we show in this work that the polynomial systems arising from an algebraic attack have a specific structure that can be exploited. Specifically, we are able to find preimages for $d$-periodic targets under the public map with a probability of $63\%$ for all security levels. The complexity of finding these preimages, as well as the fraction of $d$-periodic target increases with $d$ and hence provides a trade-off. We show that for all security levels one can choose $d=\frac{v}{2}$, for $v$ the number of vinegar variables, and reduce the security claim. Our experiments show practical running times for lower $d$ ranging from 6 seconds to 14 hours.

Meet-in-the-middle (MitM) is a powerful approach for the cryptanalysis of symmetric primitives. In recent years, MitM has led to many improved records about key recovery, preimage and collision attacks with the help of automated tools. However, most of the previous work target $\texttt{AES}$-like hashing where the linear layer is an MDS matrix. And we observe that their automatic model for MDS matrix is not suitable for primitives using a binary matrix as their linear layer. In this paper, we propose the $\texttt{n-XOR}$ model to describe the $\texttt{XOR}$ operation with an arbitrary number of inputs. And it can be applied to primitives with a binary matrix of arbitrary size. Then, we propose a check model to eliminate the possible inaccuracies caused by $\texttt{n-XOR}$. But the check model is limited by the input size (not greater than 4). Combined with the two new models, we find a MitM key recovery attack on 11-round $\texttt{Midori64}$. When the whitening keys are excluded, a MitM key recovery attack can be mounted on the 12-round $\texttt{Midori64}$. Compared with the previous best work, both of the above results have distinct advantages in terms of reducing memory and data complexity. At last, we apply the $\texttt{n-XOR}$ model to the hashing modes of primitives with large size binary matrix. The preimage attack on weakened $\texttt{camellia}-{\tt MMO}$ (without $FL/FL^{-1}$ and whitening layers) and $\texttt{Aria}-{\tt DM}$ are both improved by 1 round.

In this paper, we define the conditional constant function problem (CCFP) and, for a special case of CCFP, we propose a quantum algorithm for solving it efficiently. Such an algorithm enables us to make new evaluations to the quantum security of Feistel block cipher in the case where a quantum adversary only has the ability to make online queries in a classical manner, which is relatively realistic. Specifically, we significantly improved the chosen-plaintext key recovery attacks on two Feistel block cipher variants known as Feistel-KF and Feistel-FK. For Feistel-KF, we construct a 3-round distinguisher based on the special case of CCFP and propose key recovery attacks mounting to $r>3$ rounds. For Feistel-FK, our CCFP based distinguisher covers 4 rounds and the key recovery attacks are applicable for $r>4$ rounds. Utilizing our CCFP solving algorithm, we are able to reduce the classical memory complexity of our key recovery attacks from the previous exponential $O(2^{cn})$ to $O(1)$. The query complexity of our key recovery attacks on Feistel-KF is also significantly reduced from $O(2^{cn})$ to $O(1)$ where $c$'s are constants. Our key recovery results enjoy the current optimal complexities. They also indicate that quantum algorithms solving CCFP could be more promising than those solving the period finding problem.

Software based cryptographic implementations provide flexibility but they face performance limitations. In contrast, hardware based cryptographic accelerators utilize application-specific customization to provide real-time security solutions. Cryptographic instruction-set extensions (CISE) combine the advantages of both hardware and software based solutions to provide higher performance combined with the flexibility of atomic-level cryptographic operations. While CISE is widely used to develop security solutions, side-channel analysis of CISE-based devices is in its infancy. Specifically, it is important to evaluate whether the power usage and electromagnetic emissions of CISE-based devices have any correlation with its internal operations, which an adversary can exploit to deduce cryptographic secrets. In this paper, we propose a test vector leakage assessment framework to evaluate the pre-silicon prototypes at the early stages of the design life-cycle. Specifically, we first identify functional units with the potential for leaking information through power side-channel signatures and then evaluate them on system prototypes by generating the necessary firmware to maximize the side-channel signature. Our experimental results on two RISC-V based cryptographic extensions, RISCV-CRYPTO and XCRYPTO, demonstrated that seven out of eight prototype AES- and SHA-related functional units are vulnerable to leaking cryptographic secrets through their power side-channel signature even in full system mode with a statistical significance of $\alpha = 0.05$.

While many side-channel attacks on elliptic curve cryptography can be avoided by coordinate randomization, this is not the case for the zero-value point (ZVP) attack. This attack can recover a prefix of static ECDH key but requires solving an instance of the dependent coordinates problem (DCP), which is open in general. We design a new method for solving the DCP on GLV curves, including the Bitcoin secp256k1 curve, outperforming previous approaches. This leads to a new type of ZVP attack on multiscalar multiplication, recovering twice as many bits when compared to the classical ZVP attack. We demonstrate a $63\%$ recovery of the private key for the interleaving algorithm for multiscalar multiplication. Finally, we analyze the largest database of curves and addition formulas with over 14 000 combinations and provide the first classification of their resistance against the ZVP attack.

As the industry prepares for the transition to post-quantum secure public key cryptographic algorithms, vulnerability analysis of their implementations is gaining importance. A theoretically secure cryptographic algorithm should also be able to withstand the challenges of physical attacks in real-world environments. MAYO is a candidate in the ongoing first round of the NIST post-quantum standardization process for selecting additional digital signature schemes. This paper demonstrates three first-order single-execution fault injection attacks on a MAYO implementation in an ARM Cortex-M4 processor. By using voltage glitching to disrupt the computation of the vinegar seed during the signature generation, we enable the recovery of the secret key directly from the faulty signatures. Our experimental results show that the success rates of the fault attacks in a single execution are 36%, 82%, and 99%, respectively. They emphasize the importance of developing countermeasures against fault attacks prior to the widespread deployment of post-quantum algorithms like MAYO.

We present a key-recovery attack on a variant of the Seasign signature scheme presented by [Kim24], which attempts to avoid rejection sampling by presampling vectors $\mathbf{f}$ such that the $\mathbf{f}-\mathbf{e}$ is contained in an acceptable bound, where $\mathbf{e}$ is the secret key. We show that this choice leads to a bias of these vectors such that, in a small number of signatures, the secret key can either be completely recovered or its keyspace substantially reduced. In particular, on average, given $20$ signatures, with parameter set II of their paper, the attack reduces the private key to 128 possibilities

When designing filter functions in Linear Feedback Shift Registers (LFSR) based stream ciphers, algebraic criteria of Boolean functions such as the Algebraic Immunity (AI) become key characteristics because they guarantee the security of ciphers against the powerful algebraic attacks. In this article, we investigate a generalization of the algebraic attacks proposed by Courtois and Meier on filtered LFSR twenty years ago. We consider how the standard algebraic attack can be generalized beyond filtered LFSR to stream ciphers applying a Boolean filter function to an updated state. Depending on the updating process, we can use different sets of annihilators than the ones used in the standard algebraic attack; it leads to a generalization of the concept of algebraic immunity, and more efficient attacks. To illustrate these strategies, we focus on one of these generalizations and introduce a new notion called Extreme Algebraic Immunity (EAI). We perform a theoretic study of the EAI criterion and explore its relation to other algebraic criteria. We prove the upper bound of the EAI of an n-variable Boolean function and further show that the EAI can be lower bounded by the AI restricted to a subset, as defined by Carlet, Méaux and Rotella at FSE 2017. We also exhibit functions with EAI guaranteed to be lower than the AI, in particular we highlight a pathological case of functions with optimal algebraic immunity and EAI only n/4. As applications, we determine the EAI of filter functions of some existing stream ciphers and discuss how extreme algebraic attacks using EAI could apply to some ciphers. Our generalized algebraic attack does not give a better complexity than Courtois and Meier's result on the existing stream ciphers. However, we see this work as a study to avoid weaknesses in the construction of future stream cipher designs.

Cryptojacking, the unauthorised use of computing resources to mine cryptocurrency, has emerged as a critical threat in today’s digital landscape. These attacks not only compromise system integrity but also result in increased costs, reduced hardware lifespan, and heightened network security risks. Early and accurate detection is essential to mitigate the adverse effects of cryptojacking. This study focuses on developing a semi-supervised machine learning (ML) approach that leverages an autoencoder for feature extraction and a random forest (RF) model for classification. The objective is to enhance cryptojacking detection while maintaining a balance between accuracy and interpretability. The proposed methodology is further enhanced with explainable artificial intelligence (XAI) techniques such as local interpretable model-agnostic explanations (LIME) to offer insights into model predictions. Results from datasets such as UGRansome and BitcoinHeist indicate that the semi-supervised approach achieves accuracy rates ranging from 70% to 99%. The study demonstrates that the proposed model provides an efficient, interpretable, and scalable solution for real-time cryptojacking detection across various scenarios.

Fully Homomorphic Encryption (FHE) allows a server to perform computations directly over the encrypted data. In general FHE protocols, the client is tasked with decrypting the computation result using its secret key. However, certain FHE applications benefit from the server knowing this result, especially without the aid of the client. Providing the server with the secret key allows it to decrypt all the data, including the client's private input. Protocols such as Goldwasser et. al. (STOC'13) have shown that it is possible to provide the server with the capability of conditional decryption that allows it to decrypt the result of some pre-defined computation and nothing else. While beneficial to an honest-but-curious server to aid in providing better services, a malicious server may utilize this added advantage to perform active attacks on the overall FHE application to leak secret information. Existing security notions fail to capture this scenario since they assume that only the client has the ability to decrypt FHE ciphertexts. Therefore, in this paper, we propose a new security notion named IND-CPA$^{\text{C}}$, that provides the adversary with access to a conditional decryption oracle. We then show that none of the practical exact FHE schemes are secure under this notion by demonstrating a generic attack that utilizes this restricted decryption capability to recover the underlying errors in the FHE ciphertexts. Given the security guarantee of the underlying (R)LWE hardness problem collapses with the leakage of these error values, we show a full key recovery attack. Finally, we propose a technique to convert any IND-CPA secure FHE scheme into an IND-CPA$^\text{C}$ secure FHE scheme. Our technique utilizes Succinct Non-Interactive Argument of Knowledge, where the server is forced to generate a proof of an honest computation of the requested function along with the computation result. During decryption, the proof is verified, and the decryption proceeds only if the verification holds. Both the verification and decryption steps run inside a Garbled Circuit and thus are not observable or controllable by the server.

Sparse binary LWE secrets are under consideration for standardization for Homomorphic Encryption and its applications to private computation [20]. Known attacks on sparse binary LWE secrets include the sparse dual attack [5] and the hybrid sparse dual-meet in the middle attack [19], which requires significant memory. In this paper, we provide a new statistical attack with low memory requirement. The attack relies on some initial lattice reduction. The key observation is that, after lattice reduction is applied to the rows of a q-ary-like embedded random matrix A, the entries with high variance are concentrated in the early columns of the extracted matrix. This allows us to separate out the “hard part” of the LWE secret. We can first solve the sub-problem of finding the “cruel” bits of the secret in the early columns, and then find the remaining “cool” bits in linear time. We use statistical techniques to distinguish distributions to identify both cruel and cool bits of the secret. We recover secrets in dimensions n = 256, 512, 768, 1024 and provide concrete attack timings. For the lattice reduction stage, we leverage recent improvements in lattice reduction (flatter [34]) applied in parallel. We also apply our new attack to RLWE with 2-power cyclotomic rings, showing that these RLWE instances are much more vulnerable to this attack than LWE.

This paper presents quantum circuits for the Nguyen–Vidick (NV) sieve algorithm to solve the Shortest Vector Problem (SVP) in lattice-based cryptography. We focus on optimizing the circuit depth of the quantum NV sieve, leveraging Grover’s algorithm to reduce the search complexity. Using the proposed quantum NV sieve, we estimate the quantum re- sources required to solve SVP for various dimensions. Specifically, for a dimension size of 512 (the parameter for Kyber-512), our implemen- tation achieves a quantum attack cost of 2126.0045 in terms of the gate count–depth product metric used by National Institute of Standards and Technology (NIST). To optimize circuit depth, we employ carry-lookahead and carry-save adders for efficient multi-addition operations. Further, our quantum NV sieve performs precise sieving by implementing fixed-point arithmetic, incorporating essential components (such as input setting, up-scaling, and two’s complement). To the best of our knowledge, previous work on quantum cryptanaly- sis of SVP using the sieve algorithm has remained theoretical, without proposing quantum circuits. Our work humbly demonstrates that the post-quantum security of lattice- based cryptography (with respect to the quantum attack complexity) falls between that of multivariate-based and code-based cryptography.