In this paper, we provide details on the proofs of the quantum polynomial time algorithm of Biasse and Song (SODA 16) for computing the $S$-unit group of a number field. This algorithm directly implies polynomial time methods to calculate class groups, $S$-class groups, relative class group and unit group, ray class groups, solve the principal ideal problem, solve certain norm equations, and decompose ideal classes in the ideal class group. Additionally, combined with a result of Cramer, Ducas, Peikert and Regev (Eurocrypt 2016), the resolution of the principal ideal problem allows one to find short generators of a principal ideal. Likewise, methods due to Cramer, Ducas and Wesolowski (Eurocrypt 2017) use the resolution of the principal ideal problem and the decomposition of ideal classes to find so-called ``mildly short vectors'' in ideal lattices of cyclotomic fields.

This paper presents a security analysis of the ARX-based lightweight block cipher proposed by Mohanapriya and Nithish (Sci Rep 15, 36060 (2025)) for image encryption in IoT environments. The cipher employs a 64-bit key and a 64-bit block size, relying on Addition, Rotation, and XOR (ARX) operations. Our analysis demonstrates that the full-round version of this cipher is vulnerable to differential cryptanalysis. In fact, the cipher can be distinguished from a random permutation using $2^{41}$ chosen plaintexts. Consequently, the designers' security claims are not fully supported.

In the last decade, the introduction of advanced cryptographic protocols operating on large finite fields $\mathbb{F}_q$ has raised the need for efficient cryptographic primitives in this setting, commonly referred to as Arithmetization-Oriented (AO). The cryptanalysis of AO hash functions is essentially done through the study of the CICO problem on the underlying permutation. Two recent works at Crypto 2024 and Asiacrypt 2024 managed to solve the CICO problem much more efficiently than traditional Gröbner basis methods, using respectively advanced Gröbner basis techniques and resultants. In this paper, we propose an attack framework based on resultants that applies to a wide range of AO permutations and improves significantly upon these two recent works. Our improvements mainly come from an efficient reduction procedure that we propose and rigorously analyze, taking advantage of fast multivariate multiplication. We present the most efficient attacks on Griffin, Arion, Anemoi, and Rescue. We show that most variants of Griffin, Arion and Anemoi fail to reach the claimed security level. For the first time, we successfully break a parameter set of Rescue, namely its $512$-bit security variant. The presented theory and complexity estimates are backed up with experimental attacks. Notably, we practically find CICO solutions for $8$ out of $10$ rounds of Griffin, $11$ out of $20$ rounds of Anemoi, $6$ out of $18$ rounds of Rescue, improving by respectively $1$, $3$ and $1$ rounds on the previous best practical attacks.

The search for optimal differential trails for ARX ciphers is known to be difficult and scale poorly as the word size (and the branching through the carries of modular additions) increases.To overcome this problem, one may approximate the modular addition with the XOR operation, a process called linearization. The immediate drawback of this approach is that many valid and good trails are discarded. In this work, we explore different partial linearization trade-offs to model the modular addition through the \emph{window heuristic}, which restricts carry propagation to windows of $w_s$ consecutive positions. This strategy enables the exploration of full linearization ($w_s = 0$), normal modelling ($w_s = n$), and all the different trade-offs between completeness and speed in between. We give the corresponding SAT and MILP model and their parallel versions, and apply them to \chachacore, \speckfamily, \leafamily, and \hightfamily. Our method greatly outperforms all previous modeling of modular addition. In particular, we find the first differential path for 4 rounds of \chachacore with a probability greater than $2^{-256}$, and a corresponding 6 rounds boomerang distinguisher. This indicates that purely differential-based attacks have the potential to become competitive with differential-linear attacks, currently, the best-known attacks against \chachacore and other ARX ciphers. Finally, we exhibit an improved key recovery attack on reduced \leafamily.

Post-Quantum key encapsulation mechanisms based on the re-encryption framework of Fujisaki and Okamoto have proved very sensitive to Plaintext Checking Oracle (PCO) attacks. The first theoretic works on PCO attacks were rapidly followed by practical attacks on real implementations, notably on NIST standardized ML-KEM. The actual realization of a PCO relies on side-channel leakages that are inherently noisy ; even more so if the implementation embeds side-channel countermeasures. In this paper we tackle the often overlooked complications caused by highly noisy PCOs. We demonstrate that the impact of wrong oracle answers can be very efficiently reduced with the use of the so-called Sequential Probability Ratio Test (SPRT). This test can be seen as an elegant and natural early abort strategy on top of the commonly used approaches based on majority-voting or the likelyhood ratio test. As far as we know, this is the first use of SPRT in the context of side-channel attacks. We show that it allows to divide by a factor up to 3 the attack complexity compared to the traditional approaches. By establishing new comparisons with recently published noisy PCO attacks we emphasize that SPRT should be considered as the novel baseline for all future works in this line of research.

The Unbalanced Oil and Vinegar (UOV) construction is the foundation of several post-quantum digital signature algorithms currently under consideration in NIST's standardization process for additional post-quantum digital signature schemes. This paper introduces new single fault injection attacks against the signing procedure of deterministic variants of signature schemes based on the UOV construction. We show how these attacks can be applied to attack MAYO and PROV, two signature schemes submitted to the NIST call for additional post-quantum signature schemes. The attacks are demonstrated with reference implementations that run on an ARM Cortex-M4 processor. Our attacks do not require precise triggering or precise fault injection capabilities. Any type of fault in large portions of the code has the potential to result in successful key recovery. We demonstrate our attacks with very cheap equipment and simple clock glitching techniques, enabling the recovery of the secret key with either two faulty signatures or one correct signature and one faulty signature in the case of MAYO and one correct signature and two faulty signatures in case of PROV. The fact that our attacks do not require precise fault injection capabilities and can be successful with only a few signatures makes them particularly powerful, hence harmful for the implementation security of post-quantum digital signature schemes.

The interest in hash-based signatures (HBS) has increased since the need for post-quantum cryptography (PQC) emerged that could withstand attacks by quantum computers. Since their standardization, stateful HBS algorithms have been deployed in several products ranging from embedded devices up to servers. In practice, they are most applicable to verify the integrity and authenticity of data that rarely changes, such as the firmware of embedded devices. The verification procedure then takes place during a secure boot or firmware update process. In past works, the research community has investigated hardware and software optimizations for this use case and vendors brought forward products. In this study, we practically evaluate a fault attack on the Winternitz One-Time Signature (WOTS) scheme. The attack can be mounted on different HBS schemes, such as LMS, XMSS, and SPHINCS+. Both, the verification as well as the signing operation can be targeted. The study describes the preparation and implementation of the attack on a standard microcontroller as well as the difficulties the attacker has to overcome. Additionally it presents a countermeasure, which is easy to implement and can increase the effort for an attacker significantly.

Quasi-Abelian Syndrome Decoding (QA-SD) is a recently introduced generalization of Ring-LPN that uses multivariate polynomials rings. As opposed to Ring-LPN, it enables the use of small finite field such as GF(3) and GF(4). It was introduced by Bombar et al (Crypto 2023) in order to obtain pseudorandom correlation generators for Beaver triples over small fields. This theoretical work was turned into a concrete and efficient protocol called F4OLEage by Bombar et al. (Asiacrypt 2024) that allows several parties to generate Beaver triples over GF(2). We propose efficient algorithms to solve the decoding problem underlying the QA-SD assumption. We observe that it reduce to a sparse multivariate polynomial interpolation problem over a small finite field where the adversary only has access to random evaluation points, a blind spot in the otherwise rich landscape of sparse multivariate interpolation. We develop new algorithms for this problem: using simple techniques we interpolate polynomials with up to two monomials. By sending the problem to the field of complex numbers and using convex optimization techniques inspired by the field of “compressed sensing”, we can interpolate polynomials with more terms. This enables us to break in practice parameters proposed by Bombar et al. at Crypto’23 and Asiacrypt’24 as well as Li et al. at Eurocrypt’25 (IACR flagship conferences Grand Slam). In the case of the F4OLEage protocol, our implementation distinguishes the output of the pseudorandom correlation generator from random with advantage 60% in a few hours. This not only invalidates the security proofs, but it yields real-life privacy attacks against multiparty protocols using the Beaver triples generated by the broken pseudorandom correlation generators.

The advent of quantum computing necessitates a rigorous reassessment of classical cryptographic primitives, particularly lightweight block ciphers (LBCs) deployed in resource-constrained environments. This work presents a comprehensive quantum implementation and security analysis of the Feistel-based LBC MIBS against quantum cryptanalysis. Using the inherent reversibility of its structure, we develop a novel ancilla-free quantum circuit that optimizes qubit count and depth. For MIBS-64 and MIBS-80, our implementation achieves quantum costs of 23,371 and 24,363, requiring 128 and 144 qubits, respectively, with a depth of 4,768. We subsequently quantify the cipher's vulnerability to Grover’s key-search algorithm under the NIST PQC security constraint $\texttt{MAXDEPTH}$. By constructing Grover oracles using inner parallelization with multiple plaintext-ciphertext pairs to suppress false positives, we demonstrate total quantum attack costs of approximately $2^{94}$ for MIBS-64 and $2^{111}$ for MIBS-80. These values fall below NIST’s Level-1 security threshold ($2^{170}$), confirming the susceptibility of both MIBS variants to quantum key-recovery attacks despite their classical lightweight efficiency.

We propose a deterministic algorithm based on Coppersmith's method that employs a rank-3 lattice to address factoring-related problems. An interesting aspect of our approach is that we utilize the second vector in the LLL-reduced basis to avoid trivial collisions in the Baby-step Giant-step method, rather than the shortest vector as is commonly used in the literature. Our results are as follows: - Compared to the result by Harvey and Hittmeir (Math. Comp. 91 (2022), 1367–1379), who achieved a complexity of $O\left( \frac{N^{1/5} \log^{16/5} N}{(\log \log N)^{3/5}} \right)$ for factoring a semiprime $N = pq$, we demonstrate that in the balanced $p$ and $q$ case, the complexity can be improved to $O\left( \frac{N^{1/5} \log^{13/5} N}{(\log\log N)^{3/5}} \right).$ - For factoring sums and differences of powers, i.e., numbers of the form $N = a^n \pm b^n$, we improve Hittmeir's result (Math. Comp. 86 (2017), 2947–2954) from $O(N^{1/4} \log^{3/2} N)$ to $O\left( {N^{1/5} \log^{13/5} N} \right).$ - For the problem of finding $r$-power divisors, i.e., finding all integers $p$ such that $p^r \mid N$, Harvey and Hittmeir (Proceedings of ANTS XV, Res. Number Theory 8 (2022), no.4, Paper No. 94) recently directly applied Coppersmith's method and achieved a complexity of $O\left(\frac{N^{1/4r} \log^{10+\epsilon} N}{r^3}\right)$. By using faster LLL-type algorithm and sieving on small primes, we improve their result to $O\left(\frac{N^{1/4r} \log^{7+3\epsilon} N}{(\log\log N-\log 4r)r^{2+\epsilon}}\right)$. The worst case running time for their algorithm occurs when $N = p^r q$ with $q = \Theta(N^{1/2})$. By focusing on this case and employing our rank-3 lattice approach, we achieve a complexity of $O\left(r^{1/4} N^{1/4r} \log^{5/2} N \right)$. In conclusion, we offer a new perspective on these problems, which we hope will provide further insights.

The security of block ciphers such as AES-128, AES-192, and AES-256 relies on the assumption that their ciphertext outputs are computationally indistinguishable from random permutations. While distinguishers have been proposed for reduced-round variants or under non-standard models such as known-key or chosen-key settings, no effective distinguisher has been demonstrated for the full-round AES ciphers in the standard secret-key model. This work introduces FESLA (Feature Enhanced Statistical Learning Attack), a hybrid statistical learning framework that integrates outputs from a suite of classical statistical tests with machine learning and deep learning classifiers to construct ciphertext-only distinguishers for AES-128, AES-192, and AES-256. In contrast to existing approaches based on handcrafted or bitwise features, FESLA aggregates intermediate statistical metrics as features, enabling the capture of persistent structural biases in ciphertext distributions. Experimental evaluation across multiple datasets demonstrates consistent 100% classification accuracy using Support Vector Machines, Random Forests, Multi-Layer Perceptron, Logistic Regression, and Naïve Bayes classifiers. Generalization and robustness are confirmed through k-fold cross-validation, including on previously unseen ciphertext samples. These results establish the first ciphertext-only distinguishers for full-round AES-128, AES-192, and AES-256 under the secret-key model, and underscore the potential of machine learning–augmented cryptanalysis based on principled statistical feature engineering.

In recent years and with the emergence of the industrial revolution, the secure data sharing schemes have been developed in IoT platforms and have been recognized as a hot topic in industry and academia. These schemes enable IoT devices to securely share their sensed data in industrial environments with clients through an appropriate infrastructure and intermediary entities. The research conducted in this field shows the existence of various security challenges and solutions. Data privacy, data authentication, fairness and accountability are some of the most important security features presented. Recently, in paper [Sengupta-Ruj-Bit, TNSM 2023] proposed a secure sharing scheme and claimed that it covers all the mentioned security features even when entities collude with each other. In this paper, we investigate the security analysis of this scheme, and show that it does not cover the claimed fairness property. Therefore, the mentioned scheme is vulnerable and cannot be used as a valid scheme in real-world applications.

We present a cryptanalysis of a multi-party key exchange protocol over a modified supertropical semiring, as proposed in a recent work of R. Ponmaheshkumar, J. Ramalingam, and R. Perumal. Building on the established methods for solving linear systems $A \otimes x=b$ over the tropical semiring, as well as on our recent work on solving such systems over layered semirings such as the symmetrized and supertropical semirings, we develop a method to compute a solution of $A \otimes x=b$ over the above mentioned modified supertropical semiring. This method enables the attacker to recover the shared secret key by solving the one-sided linear system derived from the public messages of the protocol. Our findings show that this modified supertropical platform does not provide the intended security and motivate further exploration of secure semiring-based constructions.

Fault attacks have historically been one of the most popular gray-box attacks in Cryptographic literature. In such attacks, an attacker tries to inject perturbations while executing a cipher and exploit the faulty outputs to recover the key. While the efficiency of such an attack is measured by number of faults required and size of the reduced key-space, another pivotal temporal parameter in the point of fault injection which has not received considerable attention. In this work, we plug this gap for a special class of ciphers namely DEFAULT and BAKSHEESH which boast of an SBox with one or more linear structures (LS). We make new observations which lead to the improvement of Division Property based fault attacks (DIFA) introduced by Kundu et al. in ACNS 2023. We show that these linear structures are particularly responsible for giving the higher fault penetration in these ciphers. We improve the state-of-the-art for BAKSHEESH from 30 to 28 rounds and for DEFAULT from 75 to 72 using the single round random nibble fault model. While for BAKSHEESH we are able to uniquely recover the key, for DEFAULT, we are able to reduce the key-space to $2^{64}$. This leads to the best fault attacks on BAKSHEESH and DEFAULT in terms of number of rounds penetrated for fault injection. Our work reiterates the fact that a property induced (in this case LS in SBox)) for some particular Cryptographic purposes (like fault attack resistance) may manifest orthogonally for another (increasing fault penetration) and thus adds value to block cipher design space exploration and fault attack counter-measure development.

Side-channel analysis (SCA) exploits physical leakages such as power consumption or electromagnetic emissions to extract secret information from cryptographic implementations. Leakage model is the critical link between observable physical emanations (e.g., power consumption) and internal cryptographic states. When a leakage model fails to match the device’s underlying physical leakage, even powerful attacks like Correlation Power Analysis (CPA) are rendered ineffective. This fundamental challenge limits the success of traditional SCA, especially against noisy or masked targets. In this work, we introduce the Neural Leakage Model (NLM), a novel neural network architecture trained to learn and accurately characterize complex physical leakage to enhance CPA. Moving beyond NLM's superior attack performance, we directly address the critical ``black box'' problem in deep learning-based side-channel analysis (DLSCA) by proposing a novel methodology to transform the trained NLM into a mathematically equivalent, closed-form polynomial expression. This provides interpretability, offering evaluators transparent insight into the precise leakage model the network has learned. We validate NLM’s performance across four diverse datasets spanning three platforms, from low-end microcontrollers to high-end multi-core ARM systems and dedicated FPGAs. Notably, NLM successfully recovers the secret key from the highly challenging and noisy CHES2025 dataset, representing the first known successful CPA attack on this benchmark. Furthermore, NLM demonstrates superior or comparable performance when extended to higher-order CPA against masking countermeasures. Our findings establish NLM as a powerful, generalizable, and interpretable approach to modern side-channel analysis.

The sparse secret Learning with Errors (LWE) problem is a widely used assumption in efficient fully homomorphic constructions. In [Wenger et al. IEEE S\&P 2025] two approaches, `Cool and Cruel’ (C+C) and the machine learning based `SALSA', were benchmarked against the well established primal attack on sparse secrets. The authors concluded that C+C outperforms SALSA and both outperform the primal attack. In this work we show that the apparently novel C+C is an instantiation of a known dual attack [Albrecht, EUROCRYPT 2017]. To argue this we introduce a framework for dimension reduction in the bounded distance decoding problem that may be of independent interest. Furthermore we prove that the C+C `phenomenon' is an expression of the geometry of the well known Z-shape basis in $q$-ary lattices, despite claims to the contrary. We also show that a correctly parametrised primal attack outperforms C+C both in parameter regimes studied by Wenger et al. and in new parameter regimes. To support this claim, we provide an open source implementation of two variants of the primal attack that are relevant for sparse secret LWE: Drop+Solve [May--Silverman, CaLC 2001] and Guess+Verify [Albrecht et al. SAC 2019].

The security of many arithmetization-oriented (AO) hash functions depends of the hardness of Constrained-input constrained-output (CICO) problems. These problems have received significant attention from the cryptographic community in recent years, with notable advances in Gröbner basis and resultant-based attacks, yet progress has mainly been limited to CICO problems restricted to a single output. In this work, we build on the "FreeLunch method" of Bariant et al. (Crypto 2024) that constructs Gröbner bases "for free" in this particular case, and extend it to CICO problems with multiple outputs. More precisely, we consider tools for solving weighted polynomial systems, and show how to apply them in the AO setting. This results in new polynomial modelings, more efficient methods for computing the initial Gröbner basis under certain assumptions, and improved complexity estimates for the change of ordering step, derived from tighter upper bounds on the ideal degree. We apply our framework to Poseidon, Neptune and XHash8, where our assumptions are experimentally verified, and theory matches practice. For Griffin and ArionHash our assumptions are not verified, leaving us with improved, yet loose, upper bounds on the ideal degree. While our results do not threaten the security of any full-round hash function, they provide new insights into the security of these primitives under more general CICO problems.

The ARADI block cipher is developed by the U.S. National Security Agency (NSA) as part of its efforts to design secure and efficient cryptographic algorithms. In this paper, we present the first detailed analysis of the lightweight block cipher ARADI under differential fault attacks. Although ARADI is structured around word-wise operations, its security fundamentally depends on bit-level properties, making it vulnerable to carefully crafted fault models. Bit-based fault models require a large number of faults, while word-based models reduce the number of faults but make precise identification of fault locations difficult. To overcome these limitations, we introduce a new nibble-based approach using permissible nibble differences (PNDs). This model provides an effective tradeoff between the number of required faults and the complexity of identifying fault positions. In our attack, the adversary randomly injects nibble faults before the last two rounds and constructs differential equations involving unknown key variables without knowing the exact fault values. Our results demonstrate that, with at most 108 random fault injections, the entire master key of ARADI can be recovered within a practical time complexity. This work reveals previously unexplored vulnerabilities of ARADI under fault attacks and highlights the need to evaluate lightweight cipher designs using nibble-level fault models, along with the traditional bit- and word-level analyses.

Embedded devices commonly rely on digital signatures to ensure both integrity and authentication. For example, digital signatures are typically verified during the boot process or firmware updates to verify the integrity of a system. They are also used to ensure authenticity of a communication party in secure protocols. Fault injection can be used to tamper with a device in order to cause malfunctioning during cryptographic computations. For example, fault injections can be used to disturb digital signing operations. With the right type of fault an attacker can compute private keys from faulted signatures. However, fault injections can also be used during verification to get maliciously crafted digital signatures accepted during signature verification with catastrophic consequences for the security of an embedded device. In this paper, we introduce new non-obvious fault injection attacks on the verification routines of Dilithium and Falcon signature schemes, which allow an attacker to get signatures for arbitrary messages accepted by fault injection. We demonstrate the feasibility of our attacks by simulations using an ARM Cortex-M4 and the pqm4 library as a target of evaluation and pinpoint vulnerable instructions. Finally, we propose and discuss possible countermeasures against these attacks.

Deep learning-based profiled side-channel analysis (SCA) targeting cryptographic implementations has attracted significant attention in recent years. Generalizable deep learning mechanisms, such as contrastive learning-based profiled SCA (CL-SCA), can enhance the effectiveness of SCA without reliance on specific network architectures and hyperparameters. This independence enables robust adaptation across diverse attack scenarios. Nonetheless, CL-SCA relies heavily on data augmentation and may mistakenly push apart physical leakage traces that should belong to the same class, which interferes with the extraction of discriminative features crucial for SCA performance. To address these limitations, we propose a profiled SCA method based on supervised contrastive learning, called SupCL-SCA. This method enhances the learning of discriminative features that facilitate key recovery by leveraging supervised information to guide the extraction of similarities in feature space. Compared with state-of-the-art methods, SupCL-SCA not only retains their general applicability and inherent advantages but also eliminates reliance on complex data augmentation and multi-stage training. Additionally, we propose a cosine distance-based Intra-Inter Distance Ratio (IIDR) metric to assess the discriminative capability of models in deep learning-based profiled SCA methods. We evaluate SupCL-SCA on three publicly available datasets covering different implementations and platforms. Experimental results show that SupCL-SCA consistently reduces the number of traces required to recover the key compared to the original methods, demonstrating enhanced attack capability.

This work describes vulnerabilities in the specification of AEAD modes and Key Wrap in two cryptographic message formats. Firstly, this applies to AEAD packets as introduced in the novel LibrePGP specification that is implemented by the widely used GnuPG application. Secondly, we describe vulnerabilities in the AES-based AEAD schemes as well as the Key Wrap Algorithm specified in the Cryptographic Message Syntax (CMS). These new attacks exploit the possibility to downgrade AEAD or AES Key Wrap ciphertexts to valid legacy CFB- or CBC-encrypted related ciphertexts and require that the attacker learns the content of the legacy decryption result. This can happen in two principal ways: either due to the human recipient returning the decryption output to the attacker as a quote or due to a programmatic decryption oracle in the receiving system that reveals information about the plaintext. The attacks effect the decryption of low-entropy plaintext blocks in AEAD ciphertexts and, in the case of LibrePGP, also the manipulation of existing AEAD ciphertexts. For AES Key Wrap in CMS, full key decryption is possible. Some of the attacks require multiple successful oracle queries. The attacks thus demonstrate that CCA2 security is not achieved by the LibrePGP and CMS AEAD or Key Wrap encryption in the presence of a legacy cipher mode decryption oracle.

CROSS is a post-quantum secure digital signature scheme submitted to NIST’s Call for Additional Signatures which was recently selected for round 2. It features signature and key sizes in the range of SLH-DSA while providing a substantially faster signing operation. Within this work, we provide the first passive side-channel attack on the scheme. The attack recovers the secret key from all except one parameter sets from a single power trace while requiring at maximum two power traces for the R-SDP(G) 1 Fast instance. To successfully mount the attack, we show how to recover the secret key from side-channel information gained from the syndrome computation in CROSS’ identification protocol. We furthermore show how the hypothesis space for the attack can be restricted using information from the published signature.

In this note we identify major issues with “Exact Coset Sampling for Quantum Lattice Algorithms” by Zhang. The issues include: • in the first version, the proposed algorithm requires foreknowledge of the answer to compute the answer; • in the latest version, the proposed algorithm displays basic misunderstandings of quantum mechanics. We believe the existence of these issues refute the author’s claims.

In this paper, we present a new type of algebraic attack that applies to many recent arithmetization-oriented families of permutations, such as those used in Griffin, Anemoi, ArionHash, and XHash8, whose security relies on the hardness of the constrained-input constrained-output (CICO) problem. We introduce the FreeLunch approach: the monomial ordering is chosen so that the natural polynomial system encoding the CICO problem already is a Gröbner basis. In addition, we present a new dedicated resolution algorithm for FreeLunch systems of complexity lower than applicable state-of-the-art FGLM algorithms. We show that the FreeLunch approach challenges the security of fullround instances of Anemoi, Arion and Griffin. We confirm these theoretical results with experimental results on those three permutations. In particular, using the FreeLunch attack combined with a new technique to bypass 3 rounds of Griffin, we recover a CICO solution for 7 out of 10 rounds of Griffin in less than four hours on one core of AMD EPYC 7352 (2.3GHz).

In SAC'14, Biham and Carmeli presented a novel attack on DES, involving a variation of Partitioning Cryptanalysis. This was further extended in ToSC'18 by Biham and Perle into the Conditional Linear Cryptanalysis in the context of Feistel ciphers. In this work, we formalize this cryptanalytic technique for Substitution-Permutation Networks and derive several properties. A conditional approximation is then used to approximate the $inv: GF(2^8) \to GF(2^8) : x \to x^{254}$ function, which forms the only source of non-linearity in the AES. By extending the approximation to encompass the full AES round function, a linear distinguisher for 4-round AES using $2^{125.72}$ known-plaintexts is constructed; the existence of which is often understood to be impossible. We furthermore demonstrate how to recover 32 key bits directly from this distinguisher with no data or time overhead. In addition to suggesting a new approach to advancing the cryptanalysis of the AES, this result moreover demonstrates a caveat in the standard interpretation of the Wide Trail Strategy — the design framework underlying many SPN-based ciphers published in recent years.

We study the practicality of a primal hybrid lattice attack on LWE. Despite significant research efforts, the state-of-the-art in practical LWE record computations so far is the plain primal attack with Kannan's embedding. Building on an idea by Espitau and Kirchner, Bernstein proposed in 2023 an LWE hybrid attack that asymptotically outperforms the primal attack. In a nutshell, Bernstein's attack enumerates some coordinates of an LWE key and uses the sophisticated Batch-CVP {\em (Randomized) Slicer} algorithm to solve LWE in a dimension-reduced lattice. The practical implications of this attack however remain widely unclear. One of the major obstacles for judging practicality is the lack of a fast, fully functional Slicer implementation. For the first time, we provide an efficient Slicer implementation that includes all required algorithmic ingredients like locality sensitive hashing. Building on our Slicer implementation, we implement a generalization of Bernstein's algorithm. While Bernstein's attack works only for LWE, ours also applies to a more general BDD setting. Let $(\mathbf{B}, \mathbf{t})$ be a BDD instance, where the target $\mathbf{t}$ is off from the $d$-dimensional lattice $\mathcal{L}(\mathbf{B})$ by some error~$\mathbf{e}$, sampled coordinate-wise from a distribution $\mathcal{D}$. We show that for hard BDD instances, our BDD hybrid asymptotically speeds up primal's complexity of $T=2^{0.292d + o(d)}$ down to $T^{1-\mathcal{K}}$, where $\mathcal{K} \approx \big(1+\frac{H(\mathcal{D})}{0.058}\big)^{-1}$ with $H(\cdot)$ the Shannon entropy. Depending on $\mathcal{D}$, the constant $\mathcal{K}$ can be small, making practical improvements difficult. We test our Slicer-based implementation inside an implementation of our BDD hybrid lattice attack to tackle LWE instances. We choose two ternary LWE secrets with different entropies $H(\mathcal{D})$ as used in NTRU, and the centered binomial distribution as used in Kyber. For all three distributions in all tested LWE dimensions $n \in [160, 210]$, our Slicer-based implementation practically demonstrates measurable speedups over the primal attack, up to a factor of $5$. We also show that for parameters as originally suggested by Regev, the hybrid attack cannot improve over primal.

To mitigate trust concerns in the setup phase of pairing-based zk-SNARKs, the primary solution has been the sampling of the Structured Reference String (SRS) using an MPC protocol. In 2017, Bowe, Gabizon, and Miers introduced the Powers of Tau MPC protocol for sampling a universal SRS, which has since become the main SRS generation protocol for numerous practical projects. The protocol's designers showed that for a circuit with $2^{21}$ multiplication gates, verifying the universal SRS for Groth16 zk-SNARK could take $55$ minutes for a single update. However, they clarified that "the verification is not run by individual users; it is done by the coordinator and anyone who wishes to verify the transcript of the protocol after completion". This note demonstrates the importance of verifying the final SRS by either $\textit{each}$ individual end-user or $\textit{all}$ ceremony participants to mitigate potential attacks. We discuss simple attack scenarios that highlight vulnerabilities if $\textit{each}$ end-user or $\textit{all}$ participants fail to verify the final SRS. Additionally, by leveraging batching and aggregating techniques, we introduce an efficient verification algorithm for the (original) Powers of Tau protocol, substantially reducing SRS verification time and making it practical even for large-scale ceremonies. In the case of rejection, a more efficient recursive verification approach aids in identifying malicious parties more effectively. This note aims to enhance procedural understanding of SRS generation ceremonies through the Powers of Tau protocol and improve the reliability of current ceremonies against potential threats.

Singh et. al. recently uploaded a preprint describing a new hash function inspired by the Collatz Conjecture. After performing some cursory tests, the proposed function appears to be completely unsuitable for cryptographic purposes, and should not be used.

This study investigates and improves the differential cryptanalysis of the ARADI block cipher, a low-latency cryptographic system developed by the United States National Security Agency for memory encryption. The preliminary security assessment of ARADI revealed deficiencies that require additional examination. This study revisits and corrects the key recovery attack introduced by Bellini et al. (Indocrypt 2024). By correcting these mistakes, the research presents a precise 11-round key recovery method with revised attack complexities. Additionally, the analysis is expanded to introduce a key recovery attack for 12 rounds of ARADI, leaving the security margin of the cipher to only 4 rounds.

BBS/BBS+ signatures are the most promising solution to instantiate practical and lightweight anonymous credentials. They underlie standardization efforts by the W3C and the IRTF. Due to their potential for large scale deployment, it is paramount to understand their concrete security, but a number of questions have been left open by prior works. To this end, the security proofs by Au et al. (SCN '06), Camenisch et al. (TRUST '16), and Tessaro and Zhu (EUROCRYPT '23) show reductions from $q$-SDH in groups of prime order $p$, where $q$ is the number of issued signatures. However, these prior works left the possibility open that BBS/BBS+ is {\em even more secure} than what can be guaranteed by such proofs. Indeed, while the $q$-SDH assumption is subject to an attack that uses $O(\sqrt{p/q})$ group exponentiations (Cheon, EUROCRYPT '06) for several choices of $q$, no attack with a similar complexity appears to affect either of BBS+ and {\em deterministic} BBS, for which the best known attacks amount to recovering the secret key by breaking the discrete logarithm problem. The assumption that this attack is best possible also seemingly justifies the choice of parameters in practice. Our result shows that this expectation is not true. We show new attacks against BBS+ and "deterministic" BBS which, after seeing $q$ signatures, allow us to recover the secret key with the same complexity as solving the $\Theta(q)$-Discrete Logarithm problem, which in turn is proportional to $O(\sqrt{p/q})$ for many choices of $q$. Further, we also extend the attack to a reduction showing that the security of BBS+ and deterministic BBS implies the $\Theta(q)$-SDH assumption.

Fault injection attacks (FIA) pose significant security threats to embedded systems as they exploit weaknesses across multiple layers, including system software, instruction set architecture (ISA), microarchitecture, and physical hardware. Early detection and understanding of how physical faults propagate to system-level behavior are essential to safeguarding cyberinfrastructure. This work introduces CRAFT, a framework that combines pre-silicon analysis with post-silicon validation to systematically uncover and analyze fault injection vulnerabilities. Our study, conducted on a RISC-V soft-core processor (cv$32$e$40$x) reveals two novel vulnerabilities. First, we demonstrate a method to induce instruction skips by glitching the clock (single-glitch attack), which prevents critical values from being loaded from memory, thus disrupting program execution. Second, we show a technique that converts a fetched legal instruction into an illegal one mid-execution, diverting control flow in a manner exploitable by attackers. Notably, we identified a specific timing window in which the processor fails to detect these illegal control-flow diversions, allowing silent, undetected corruption of the program state. By simulating $9248$ FIA scenarios at pre-silicon and conducting root-cause analysis of the RISC-V pipeline, we trace the faults to a previously unreported vulnerability in a pipeline register shared between the instruction fetch and decode stages. Our approach reduced the search space for post-silicon experiments by $97.31$\%, showing pre-silicon advantages for post-silicon testing. Finally, we validate our identified exploit cases on real hardware (FPGA).

The machine learning problem of model extraction was first introduced in 1991 and gained prominence as a cryptanalytic challenge starting with Crypto 2020. For over three decades, research in this field has primarily focused on ReLU-based neural networks. In this work, we take the first step towards the cryptanalytic extraction of PReLU neural networks, which employ more complex nonlinear activation functions than their ReLU counterparts. We propose a raw output-based parameter recovery attack for PReLU networks and extend it to more restrictive scenarios where only the top-m probability scores are accessible. Our attacks are rigorously evaluated through end-to-end experiments on diverse PReLU neural networks, including models trained on the MNIST dataset. To the best of our knowledge, this is the first practical demonstration of PReLU neural network extraction across three distinct attack scenarios.

The NIST Post-Quantum Cryptography (PQC) standardization has entered its fourth round, underscoring the critical importance of addressing side-channel attacks (SCA), a dominant threat in real-world cryptographic implementations, especially on embedded devices. This paper presents a novel chosen-ciphertext side-channel attack against CRYSTALS-Kyber (standardized as ML-KEM) implementations with Fisher-Yates shuffled polynomial reduction. We propose an efficient and fault-tolerant key recovery algorithm that, by crafting malicious ciphertexts, induces changes in the Hamming weight distribution of an intermediate polynomial's coefficients (the output of the shuffled polynomial reduction during decapsulation), enabling recovery of secret key coefficients from these changes. To ensure robustness, we propose an error-correction strategy that leverages the Hamming weight classifier's behavior to constrain and shrink the correction search space, maintaining effectiveness even with less accurate classifiers or in low-SNR environments. A Multi-Layer Perceptron (MLP) is employed for Hamming weight classification from side-channel traces, achieving 97.11% accuracy. We combine statistical analysis with explainable deep learning for precise trace segmentation during pre-processing. Experimental results demonstrate full key recovery with only an average of $10 + 354 \times 3$ ciphertext queries and a success rate of 97.98%, reducing the adversarial effort by 95.36% compared to contemporary bit-flip techniques. Although shuffling aims to disrupt temporal correlations, our results show that statistical features persist and leak through shuffled implementations. This work reveals enduring SCA risks in shuffled implementations and informs a broader reassessment of PQC side-channel resilience.

Masking is one of the most widespread countermeasures against side-channel attacks. However, its integration into hardware implementation is subject to physical hazards that can mitigate its security. To counter glitches, the most studied physical hazard, an effective solution is to resynchronize the signals by integrating additional hardware layers into the architecture. However, these solutions have an impact on the performance of the implementation. A solution to avoid these limitations is to use more shares to compute higher-degree functions. We study the cost of this approach, denominated ($td+n$)-masking. We first derive optimal dependence structures for the creation on non-complete sharings, which allow us to obtain efficient implementation of substitution boxes. As a case study, we use these elements to create a second-order masked architecture for the PRESENT cipher. We perform multiple TVLA tests to validate our approach. Our results confirm that the strategy is efficient in terms of performance, at the cost of increased hardware resources.

We present a new distinguisher for alternant and Goppa codes, whose complexity is subexponential in the error-correcting capability, hence better than that of generic decoding algorithms. Moreover it does not suffer from the strong regime limitations of the previous distinguishers or structure recovery algorithms: in particular, it applies to the codes used in the Classic McEliece candidate for postquantum cryptography standardization. The invariants that allow us to distinguish are graded Betti numbers of the homogeneous coordinate ring of a shortening of the dual code. Since its introduction in 1978, this is the first time an analysis (in the CPA model) of the McEliece cryptosystem breaks the exponential barrier.

At CHES 2017, Banik et al. proposed a lightweight block cipher GIFT consisting of two versions GIFT-64 and GIFT-128. Recently, there are lots of authenticated encryption schemes that adopt GIFT-128 as their underlying primitive, such as GIFT-COFB and HyENA. To promote a comprehensive perception of the soundness of the designs, we evaluate their security against differential-linear cryptanalysis. For this, automatic tools have been developed to search differential-linear approximation for the ciphers based on S-boxes. With the assistance of the automatic tools, we find 13-round differential-linear approximations for GIFT-COFB and HyENA. Based on the distinguishers, 18-round key-recovery attacks are given for the message processing phase and initialization phase of both ciphers. Moreover, the resistance of GIFT-64/128 against differential-linear cryptanalysis is also evaluated. The 12-round and 17-round differential-linear approximations are found for GIFT-64 and GIFT-128 respectively, which lead to 18-round and 19-round key-recovery attacks respectively. Here, we stress that our attacks do not threaten the security of these ciphers.

The asymmetric cryptographic constructions upon on number- theoretic hardness assumptions have become insecure, due to Shor’s quantum algorithm and they will be vulnerable to large scale quantum computers. Hence, the adaption to quantum-resistant cryptosystems is a major task. Digital signatures, being a fundamental primitive in nu- merous applications. Recently, a new approach by Nguyen et al. [9] has claimed post-quantum security by basing the signature algorithm’s se- curity on a variant of the discrete logarithm problem. In this paper, we present a cryptanalysis of this construction and demonstrate a practi- cal forgery attack that allows generating an unlimited number of valid signatures—without access to a signing oracle.

Poseidon and Poseidon2 are cryptographic hash functions crafted for efficient zero-knowledge proof systems and have seen wide adoption in practical applications. We introduce the use of the Graeffe transform in univariate polynomial solving within this line of work. The proposed method streamlines the root recovery process in interpolation attacks and achieves several orders of magnitude acceleration in practical settings, enabling a new and more efficient class of attacks against Poseidon targeting round-reduced permutations and constrained input/output instances. We release open-source code and describe our method in detail, demonstrating substantial improvements over prior approaches: reductions in wall time by a factor of $2^{13}$ and in memory usage by a factor of $2^{4.5}$. Memory-access costs for NTTs turn out to be a dominant barrier in practice. And we prove that this cost increases at least as the $4/3$-power of the input size (up to logarithmic factors), which suggests the commonly used pseudo-linear cost model may underestimate the true resource requirements. This behavior contrasts with multivariate equation solving, whose main bottleneck remains finite-field linear algebra. We argue that, when selecting parameters, designers should account for interpolation-based attacks explicitly, since their practical hardness is determined by different, and sometimes stronger, resource constraints than those of multivariate techniques.

We show that the authentication scheme (IEEE Internet Things J., 24310-24322, 2024) cannot be practically implemented, because it misused the elliptic curve group law---point multiplication, which is represented as $kP$, where $k$ is an integer and $P$ is a point on an elliptic curve. But the scheme uses the false representation $Q_iQ_j$ to construct verification equations, where $Q_i$ and $Q_j$ are two points. Besides, we find that an adversary can retrieve the target relay device's secret key using the intercepted message via open channels.

A cryptographic primitive based on the Learning With Errors (LWE) problem with variants is a promising candidate for the efficient quantum-resistant public key cryptosystem. As the parameters for such cryptosystems are chosen by the concrete attack cost for the corresponding LWE problem, improving the LWE solving algorithm has a significant importance. In this paper, we present a new hybrid attack on the LWE problem. This new attack combines the primal lattice attack and an improved variant of the MitM attack, Meet-LWE, suggested by May (Crypto'21). This resolves the major open problem posed in the same paper. To this end, we develop several technical tools for hybrid attacks; a new property of Babai's nearest plane algorithm with respect to projection, an approximate variant of Meet-LWE, and a locality-sensitive hashing-based near-collision finding algorithm. We also present a comprehensive analysis of the proposed attack, which involves the complicated arguments of both lattice and representation techniques. We finally estimate the concrete cost of our attack for sparse LWE keys. For some currently deployed parameter sets in fully homomorphic encryption libraries, our attack achieves up to $10$ bits improvement than the previous primal hybrid attacks.

MQOM is one of the fourteen remaining candidates in the second round of the NIST post-quantum signature standardization process. Introduced in 2023, MQOM instantiates the Multi-Party Computation in the Head (MPCitH) paradigm over the well-established hard problem of solving Multivariate Quadratic (MQ) equations. In this paper, we present the first fault attacks on MQOM targeting the MQ evaluation phase, which is a central component of the algorithm. We introduce four differential fault attacks and demonstrate their effectiveness against both unprotected and masked implementations. The first two target the secret key using a random fault model, making them particularly realistic and practical. With as little as one or two injected faults, depending on the variant, the entire secret key can be recovered through linear algebra. The other two attacks exploit faults on the coefficients of the MQ system directly. Our results highlight that the MQ evaluation, despite not being identified as a sensitive component until now, can be exploited using just a few fault injections.

The ZUC stream cipher is integral to modern mobile communication standards, including 4G and 5G, providing secure data transmission across global networks. Recently, Dutta et al. (Indocrypt, 2024) presented the first quantum resource estimation of ZUC under Grover's search, Although preliminary, this work marks the beginning of quantum security analysis for ZUC. In this paper, we present an improved quantum resource estimation for ZUC, offering tighter bounds for Grover-based exhaustive key search. Beyond traditional quantum resource estimations, we also provide a concrete timescale required to execute such attacks using the specified quantum resources. Our findings show that a full-round, low depth implementation of ZUC-128 can be realized with a maximum of $375$ ancilla qubits, a T-count of $106536$, and a T-depth of $15816$. Furthermore, the Grover-based key search can be performed most efficiently using $1201$ logical qubits, $170681$ T gates, and a T-depth of $78189$, resulting in a runtime of $1.78\times10^{11}$ years, an improvement of 93.43% over the estimated $2.71 \times 10^{12}$ years by the implementation given by Dutta et al., we also provide akin analysis for ZUC-256 with an 99.23% decrease in time. These estimations are done assuming state-of-the-art superconducting qubit error-correcting technology.

Verifiable Homomorphic Encryption (VHE) is a cryptographic technique that integrates Homomorphic Encryption (HE) with Verifiable Computation (VC). It serves as a crucial technology for ensuring both privacy and integrity in outsourced computation, where a client sends input ciphertexts $\mathsf{ct}$ and a function $f$ to a server and verifies the correctness of the evaluation upon receiving the evaluation result $f(\mathsf{ct})$ from the server. Chatel et al. (CCS'24) introduced two VHE schemes: Replication Encoding (REP) and Polynomial Encoding (PE). A similar approach to REP was used by Albrecht et al. (EUROCRYPT'24) to develop a Verifiable Oblivious PRF scheme (vADDG). A key approach in these schemes is to embed specific secret information within ciphertexts and use them to verify homomorphic evaluations. This paper presents efficient forgery attacks against the verifiability guarantees of these VHE schemes. We introduce two attack strategies. The first targets REP and vADDG, extracting secret information in encrypted form from input ciphertexts and leveraging it to forge output ciphertexts without being detected by the verification algorithm. The second targets PE, exploiting its secret embedding structure to forge output ciphertexts that remain valid on input values for verification, yet violate the verifiability property. Our forgery attack on vADDG demonstrates that the proposed 80-bit security parameters provide at most 10 bits of concrete security. Our attack on REP and \PE achieves a probability 1 attack with linear time complexity when using fully homomorphic encryption.

Group actions have emerged as a powerful framework in post-quantum cryptography, serving as the foundation for various cryptographic primitives. The Lattice Isomorphism Problem (LIP) has recently gained attention as a promising hardness assumption for designing quantum-resistant protocols. Its formulation as a group action has opened the door to new cryptographic applications, including a commitment scheme and a linkable ring signature. In this work, we analyze the security properties of the LIP group action and present new findings. Specifically, we demonstrate that it fails to satisfy the weak unpredictability and weak pseudorandomness properties when the adversary has access to as few as three and two instances with the same secret, respectively. This significantly improves upon prior analysis by Budroni et al. (PQCrypto 2024). As a direct consequence of our findings, we reveal a vulnerability in the linkable ring signature scheme proposed by Khuc et al. (SPACE 2024), demonstrating that the hardness assumption underlying the linkable anonymity property does not hold.

The Linear Code Equivalence ($\mathsf{LCE}$) problem asks, for two given linear codes $\mathcal{C}, \mathcal{C}'$, to find a monomial $\mathbf{Q}$ mapping $\mathcal{C}$ into $\mathcal{C}'$. Algorithms solving $\mathsf{LCE}$ crucially rely on a (heuristic) subroutine, which recovers the secret monomial from $\Omega(\log n)$ pairs of codewords $(\mathbf{v}_i, \mathbf{w}_i)\in \mathcal{C} \times \mathcal{C}'$ satisfying $\mathbf{w}_i = \mathbf{v}_i\mathbf{Q}$. We greatly improve on this known bound by giving a constructive algorithm that provably recovers the secret monomial from any \emph{two} pairs of such codewords for fields of sufficiently large prime order $q$. Furthermore, we demonstrate that in practice this algorithm remains effective even for small prime fields. We then show that this reduction in the number of required pairs enables the design of a more efficient algorithm for solving the $\mathsf{LCE}$ problem. Our asymptotic analysis shows that this algorithm outperforms previous approaches for a wide range of parameters. Our concrete analysis reveals significant improvements as well. Most notably, in the context of the LESS signature scheme, a second-round contender in the ongoing NIST standardization effort, we achieve bit security reductions of up to 24 bits.

Blind signatures (BS) allow a signer to produce a valid signature on a message without learning the message itself. They have niche applications in privacy-preserving protocols such as digital cash and electronic voting. Non-interactive blind signatures (NIBS) remove the need for interaction between the signer and the user. In the post-quantum era, lattice-based NIBS schemes are studied as candidates for long-term security. In Asiacrypt 2024, Baldimtsi et al. proposed the first lattice-based NIBS construction, whose security relies on the random one-more inhomogeneous short integer solution (rOM-ISIS) assumption. This rOM-ISIS is considered to be a non-standard assumption. Later, Zhang et al. introduced another lattice-based construction in ProvSec 2024, and proved its security under the standard module short integer solution (MSIS) assumption. We analyse the security of the latter scheme. In the random oracle model, we show that it fails to achieve both nonce blindness and receiver blindness. We present explicit attacks where an adversary breaks both properties with probability~1. Our attack is based on a crucial observation that uncovers a flaw in the design. Specifically, this flaw allows an attacker to link a message-signature pair with its presignature-nonce pair. In addition, we also identify a flaw in the unforgeability proof. Finally, we suggest a modification to address the issue, which is similar to Baldimtsi et al. construction, and its security relies again on the non-standard rOM-ISIS assumption. This work again raises the question of the feasibility of achieving NIBS from standard assumptions.

Fiat-Shamir signatures replace the challenge in interactive identification schemes by a hash value over the commitment, the message, and possibly the signer's public key. This construction paradigm is well known and widely used in cryptography, for example, for Schnorr signatures and CRYSTALS-Dilithium. There is no ``general recipe'' for constructing Fiat-Shamir signatures regarding the inputs and their order for the hash computation, though, since the hash function is usually modeled as a monolithic random oracle. In practice, however, the hash function is based on the Merkle-Damgård or the sponge design. Our work investigates whether there are advisable or imprudent input orders for hashing in Fiat-Shamir signatures. We examine Fiat-Shamir signatures with plain and nested hashing using Merkle-Damgård or sponge-based hash functions. We analyze these constructions in both classical and quantum settings. As part of our investigations, we introduce new security properties following the idea of quantum-annoyance of Eaton and Stebila (PQCrypto 2021), called annoyance for user exposure and signature forgeries. These properties ensure that an adversary against the hash function cannot gain a significant advantage when attempting to extend a successful attack on a single signature forgery to multiple users or to multiple forgeries of a single user. Instead, the adversary must create extra forgeries from scratch. Based on our analysis, we derive a simple rule: When using Fiat-Shamir signatures, one should hash the commitment before the message; all other inputs may be ordered arbitrarily.

LOL (League of Legends) is a general framework for designing blockwise stream ciphers to achieve higher software performance in 5G/6G. Following this framework, stream ciphers LOL-MINI and LOL-DOUBLE are proposed, each with a 256-bit key. This paper focuses on analyzing their security against correlation attacks. Additionally, we propose an observation on constructing linear approximation trails. The repeated linear approximations in trails may not only increase the number of active S-boxes but also make the correlations of corresponding approximations zero. We propose a new rule to address this situation and develop multiple correlation attacks. Furthermore, we use theoretical cryptanalysis to lower the time complexity of aggregation and construct multiple linear approximations efficiently. For LOL-MINI, we construct approximately $2^{32.17}$ linear approximations with absolute correlations ranging from $2^{-118.38}$ to $2^{-127}$. We launch the developed multiple correlation attack on LOL-MINI with a time complexity of $2^{250.56}$, a data complexity of $2^{250}$, a memory complexity of $2^{250}$ and a success probability of 99.4$\%$. For LOL-DOUBLE, there similarly exist approximately ${{2}^{22.86}}$ linear approximations with absolute correlations ranging from $2^{-147.23}$ to $2^{-156}$. It is worth noting that our results do not violate the declaration of LOL-MINI since the maximum length of keystreams for a single pair of key and IV is less than $2^{64}$.

Quantum algorithms for factoring, finite-field discrete logarithms (DLP), and elliptic-curve discrete logarithms (ECDLP) are usually presented as three separate attack pipelines, even though all three are instances of the Abelian Hidden Subgroup Problem (AHSP). We prove that the semi-agnostic elliptic-curve discrete logarithm problem (semi-agnostic ECDLP), defined on smooth elliptic curves, singular nodal cubics, and over rings such as $\mathbb{Z}_N$, is complete for all standard abelian hidden subgroup problems (SAHSPs). In other words, factoring, finite-field DLP, and ECDLP all reduce to semi-agnostic ECDLP. This gives the first completeness theorem for the cryptographic subclass of HSP. To argue minimality from the practical point of view (for example, the minimal size of Shor's circuit implementation), we formalize the No Efficient Injective Homomorphism (NEIH) hypothesis: no generic polynomial-time injective homomorphism exists from elliptic curve subgroups into small cyclic groups. NEIH is strongly supported by three provable limitations: (i) the embedding-degree barrier, (ii) an algebraic rigidity lemma, and (iii) a generic-group model barrier, ruling out non-algebraic embeddings without already solving the ECDLP. The completeness result has practical implications, as it identifies a minimal universal Shor engine: a single, programmable Weierstrass circuit that implements reversible field arithmetic and point addition on a smooth or nodal Weierstrass model. This one circuit instance suffices to execute Shor's algorithm for factoring, finite-field DLP, and ECDLP without recompilation and with the same asymptotic resources as specialized designs.

Let $\star: G \times X \rightarrow X$ be the action of a group $G$ of size $N=|G|$ on a set $X$. Let $y = g \star x \in X$ be a group action dlog instance, where our goal is to compute the unknown group element $g \in G$ from the known set elements $x,y \in X$. The Galbraith-Hess-Smart (GHS) collision finding algorithm solves the group action dlog in $N^{\frac 1 2}$ steps with polynomial memory. We show that group action dlogs are suitable for precomputation attacks. More precisely, for any $s,t$ our precomputation algorithm computes within $st$ steps a hint of space complexity $s$, which allows to solve any group action dlog in an online phase within $t$ steps. A typical instantiation is $s=t=N^{\frac 1 3}$, which gives precomputation time $N^{\frac 2 3}$ and space $N^{\frac 1 3}$, and online time only $N^{\frac 1 3}$. Moreover, we show that solving multiple group action dlog instances $y_1, \ldots , y_m$ allows for speedups. Namely, our collision finding algorithm solves $m$ group action dlogs in $\sqrt{m}N^{\frac 1 2}$ steps, instead of the straight-forward $mN^{\frac 1 2}$ steps required for running $m$ times GHS. Interestingly, our multi instance algorithm (with precomputation) can be seen as a special case of our precomputation algorithm. Our multiple instance approach can be freely combined with our precomputations, allowing for a variety of tradeoffs. Technically, our precomputation and multiple instance group action dlog attacks are adaptations of the techniques from the standard dlog setting in abelian groups. While such an adaptation seems natural, it is per se unclear which techniques transfer from the dlog to the more general group action dlog setting, for which $X$ does not offer a group structure. Our algorithms have direct implications for all group action based cryptosystems, such as CSIDH and its variants. We provide experimental evidence that our techniques work well in the CSIDH setting.

We introduce the use of machine learning in the cryptanalysis of code-based cryptography. Our focus is on distinguishing problems related to the security of NIST round-4 McEliece-like cryptosystems, particularly for Goppa codes used in ClassicMcEliece and Quasi-Cyclic Moderate Density Parity-Check (QC-MDPC) codes used in BIKE. We present DeepDistinguisher, a new algorithm for distinguishing structured codes from random linear codes that uses a transformer. The results show that the new distinguisher achieves a high level of accuracy in distinguishing Goppa codes, suggesting that their structure may be more recognizable by AI models. Our approach outperforms traditional attacks in distinguishing Goppa codes in certain settings and does generalize to larger code lengths without further training using a puncturing technique. We also present the first distinguishing results dedicated to MDPC and QC-MDPC codes.

Algebraic Immunity (AI) is a fundamental property in the security analysis of stream ciphers and Boolean functions, measuring the resistance of a function against algebraic attacks. In this work, we focus on the computation of AI and, more specifically, on its generalization to restricted algebraic immunity, which considers annihilators constrained to specific subsets of $\mathbb{F}_2^n$. While the computation of AI has been studied using methods based on Reed-Muller codes and iterative rank-based algorithms, these approaches have not been formally adapted to the restricted setting. We address this gap by establishing the theoretical foundations required for the adaptation of these techniques and providing explicit algorithms for computing restricted algebraic immunity. To assess the efficiency of our algorithms, we conduct practical experiments comparing the computational cost of the Reed-Muller and iterative approaches in terms of time and memory. As a case study, we analyze the algebraic immunity restricted to the slices of $\mathbb{F}_2^n$, i.e. the sets of binary vectors of fixed Hamming weight, denoted $\mathsf{AI}_k$. The slices are of particular interest in areas of cryptography, such as side-channel analysis and stream cipher design, where the input weight may be partially predictable. We further investigate restricted algebraic immunity for Weightwise (Almost) Perfectly Balanced (W(A)PB) functions, which have been extensively studied since $2017$. Our results include an empirical analysis of $\mathsf{AI}_k$ distributions for WPB functions with $4$, $8$, and $16$ variables, as well as an evaluation of $\mathsf{AI}_k$ for various known function families.

The transition to post-quantum cryptography (PQC) is accelerating due to the potential of quantum computing to compromise classical public-key cryptosystems. While standardized schemes such as CRYSTALS-Kyber, CRYSTALS-Dilithium, and SPHINCS+ offer strong theoretical security, practical deployments remain susceptible to physical-layer vulnerabilities, notably side-channel attacks (SCAs). SCAs exploit unintentional leakages in hardware and software implementations—such as power traces, electromagnetic emissions, and timing variations—to recover secret keys without altering the target system. These attacks are non-invasive, cost-effective, and applicable across diverse platforms, making them a critical threat vector for PQC in embedded and resource-constrained environments. This survey provides a structured, in-depth review of SCAs targeting PQC implementations, encompassing both classical methods—such as Simple Power Analysis, Differential Power Analysis, Correlation Power Analysis, Template Attacks, and Mutual Information Analysis—and emerging machine learning (ML)-driven approaches. Special attention is given to deep learning models, including CNNs, RNNs, and MLPs, which have demonstrated superior performance in profiling attacks by automatically learning leakage patterns from high-dimensional trace data, even in the presence of countermeasures like masking and desynchronization. We categorize and compare recent attack strategies, analyze their effectiveness against various PQC schemes, and examine the limitations of existing countermeasures. Finally, we identify open research challenges and outline hybrid defense strategies that integrate classical protections with adaptive, ML-aware mitigation techniques. This comprehensive synthesis aims to bridge the gap between PQC algorithm design and secure, implementation-level deployment in the quantum era.

Fault injection attacks (FIAs) present a significant threat to the integrity of deep neural networks (DNNs), particularly in hardware-accelerated deployments on field-programmable gate arrays (FPGAs). These attacks intentionally introduce faults into the system, leading the DNN to generate incorrect outputs. This work presents the first successful targeted misclassification attack against a convolutional neural network (CNN) implemented on FPGA hardware, achieved by injecting a single clock glitch at the final layer (argmax) to manipulate the predicted output class. Our attack targets the commonly adopted argmax layer, a lightweight replacement for softmax in resource-constrained implementations. By precisely injecting a single clock glitch during the comparison phase of the argmax operation, the attack reliably induces misclassifications, forcing the model to 'skip' a specifically chosen class and output an incorrect label for it without affecting the computed scores of other classes. Unlike prior works that only cause random misclassifications, our attack achieves a high success rate of 80–87% for a targeted class, without inducing collateral misclassifications of other classes. Our evaluations show a significant reduction in classification accuracy, with the model’s performance dropping from an initial 94.7% to an average final accuracy ranging from 7.7–14.7%. Our attack is demonstrated on a CNN model implemented using a common systolic array architecture, which is well-suited for resource-constrained edge devices and artificial intelligence (AI) accelerators. Our study confirms the vulnerability of hardware-accelerated machine learning systems to low-cost physical attacks, emphasizing the critical need for hardware-level countermeasures in safety-critical machine learning applications.

With the increasing demand for Bluetooth devices, various Bluetooth devices support a power-saving mode to reduce power consumption. One of the features of the power-saving mode is that the Bluetooth sessions among devices are temporarily disconnected or are close to being disconnected. Prior works have analyzed that the power-saving mode is vulnerable to denial of sleep (DoSL) attacks that interfere with the transition to the power-saving mode of Bluetooth devices, thereby increasing its power consumption. However, to the best of our knowledge, no prior work has analyzed vulnerabilities or attacks on the state after transitioning to the power-saving mode. To address this issue, we present an attack that abuses two novel vulnerabilities in sleep mode, which is one of the Bluetooth power-saving modes, to break Bluetooth sessions. We name the attack Breaktooth. The attack is the first to abuse the vulnerabilities as an entry point to hijack Bluetooth sessions between victims. The attack also allows overwriting the link key between the victims using the hijacked session, enabling arbitrary command injection on the victims. Furthermore, while many prior attacks assume that attackers can forcibly disconnect the Bluetooth session using methods such as jamming to launch their attacks, our attack does not require such assumptions, making it more realistic. In this paper, we present the root causes of the Breaktooth attack and their impact. We also provide the technical details of how attackers can secretly detect the sleep mode of their victims. The attackers can easily recognize the state of the victim's Bluetooth session remotely using a standard Linux command. Additionally, we develop a low-cost toolkit to perform our attack and confirm the effectiveness of our attack. Then, we evaluate the attack on 17 types of commodity Bluetooth keyboards, mice and audio devices that support the sleep mode and show that the attack poses a serious threat to Bluetooth devices supporting the sleep mode. To prevent our attack, we present defenses and their proof-of-concept. We responsibly disclosed our findings to the Bluetooth SIG. We also released the toolkit as open-source.

Side-Channel Attacks target the recovery of key material in cryptographic implementations by measuring physical quantities such as power consumption during the execution of a program. Simple Power Attacks consist in deducing secret information from a trace using a single or a few samples, as opposed to differential attacks which require many traces. Software cryptographic implementations now all contain a data-independent execution path, but often do not consider variations in power consumption associated to data. In this work, we show that a technique commonly used to select a value from different possible values in a control-independant way leads to significant power differences depending on the value selected. This difference is actually so important that a single sample can be considered for attacking one condition, and no training on other traces is required. We exploit this finding to propose the first single-trace attack without any knowledge gained on previous executions, using trace folding. We target the two modular exponentiation implementations in Libgcrypt, getting respectively 100% and 99.98% of correct bits in average on 30 executions using 2,048-bit exponents. We also use this technique to attack the scalar multiplication in ECDSA, successfully recovering all secret nonces on 1,000 executions. Finally, the insights we gained from this work allow us to show that a proposed counter-measure from the litterature for performing the safe loading of precomputed operands in the context of windowed implementations can be attacked as well.

HuFu is an unstructured lattice-based signature scheme proposed during the NIST PQC standardization process. In this work, we present a side-channel analysis of HuFu's reference implementation. We first exploit the multiplications involving its two main secret matrices, recovering approximately half of their entries through a non-profiled power analysis with a few hundred traces. Using these coefficients, we reduce the dimension of the underlying LWE problem, enabling full secret key recovery with calls to a small block-sized BKZ. To mitigate this attack, we propose a countermeasure that replaces sensitive computations involving a secret matrix with equivalent operations derived solely from public elements, eliminating approximately half of the identified leakage and rendering the attack unfeasible. Finally, we perform a non-profiled power analysis targeting HuFu's Gaussian sampling procedure, recovering around 75\% of the remaining secret matrix's entries in a few hundred traces. While full key recovery remains computationally intensive, we demonstrate that partial knowledge of the secret significantly improves the efficiency of signature forgery.

Most concrete analyses of lattice reduction focus on the BKZ algorithm or its variants relying on Shortest Vector Problem (SVP) oracles. However, a variant by Li and Nguyen (Cambridge U. Press 2014) exploits more powerful oracles, namely for the Densest rank-$k$ Sublattice Problem ($DSP_k$) for $k \geq 2$. We first observe that, for random lattices, $DSP_2$ --and possibly even $DSP_3$-- seems heuristically not much more expensive than solving SVP with the current best algorithm. We indeed argue that a densest sublattice can be found among pairs or triples of vectors produced by lattice sieving, at a negligible additional cost. This gives hope that this approach could be competitive. We therefore proceed to a heuristic and average-case analysis of the slope of $DSP_k$-BKZ output, inspired by a theorem of Kim (Journal of Number Theory 2022) which suggests a prediction for the volume of the densest rank-$k$ sublattice of a random lattice. Under this heuristic, the slope for $k=2$ or $3$ appears tenuously better than that of BKZ, making this approach less effective than regular BKZ using the "Dimensions for Free'' of Ducas (EUROCRYPT 2018). Furthermore, our experiments show that this heuristic is overly optimistic. Despite the hope raised by our first observation, we therefore conclude that this approach appears to be a No-Go for cryptanalysis of generic lattice problems.

Neural networks have been applied to symmetric cryptanalysis, and Gohr demonstrated that a neural-network-based distinguisher achieves higher accuracy than classical differential distinguishers at CRYPTO 2019. In this work, we analyze ciphertext data through the lens of probability distributions, identifying non-random features that provide indirect insights into how neural networks distinguish ciphertexts from random data. In parallel, we improve the key-recovery attack process by adopting the Bayesian-UCB method, which achieves a better balance between exploration and exploitation of ciphertext structures. These enhancements reduce the runtime of key-recovery attacks while simultaneously increasing their success rate.

Searching for collisions in random functions is a fundamental computational problem, with many applications in symmetric and asymmetric cryptanalysis. When one searches for a single collision, the known quantum algorithms match the query lower bound. This is not the case for the problem of finding multiple collisions, despite its regular appearance as a sub-component in sieving-type algorithms. At EUROCRYPT 2019, Liu and Zhandry gave a query lower bound $\Omega{}(2^{m/3 + 2k/3})$ for finding $2^k$ collisions in a random function with $m$-bit output. At EUROCRYPT 2023, Bonnetain et al. gave a quantum algorithm matching this bound for a large range of $m$ and $k$, but not all admissible values. Like many previous collision-finding algorithms, theirs is based on the MNRS quantum walk framework, but it chains the walks by reusing the state after outputting a collision. In this paper, we give a new algorithm that tackles the remaining non-optimal range, closing the problem. Our algorithm is tight (up to a polynomial factor) in queries, and also in time under a quantum RAM assumption. The idea is to extend the chained walk to a regime in which several collisions are returned at each step, and the "walks" themselves only perform a single diffusion layer.

ChaCha is a well-known ARX-based cipher and has become one of the most widely used ciphers in the real world. In this paper, a systematic three-case framework called \emph{Mixderive} to find linear approximations for ChaCha is proposed. By this new framework, new linear approximations for 3.5- and 4-round ChaCha are found, which are significantly better than the existing linear approximations proposed at EUROCRYPT 2021 and ASIACRYPT 2022. These improvements confirm the effectiveness of \emph{Mixderive}. In addition, new 2- and 2.5-round linear approximations for ChaCha are found by \emph{Mixderive}. Based on these new findings, new differential-linear distinguishers for 7- and 7.5-round ChaCha256 with complexities ${2^{162.28}}$ and ${2^{247.08}}$ are proposed, which improve the best known distinguishers by factors of ${2^{4.61}}$ and ${2^{4.46}}$, respectively. To the best of our knowledge, both cryptanalytic results are the best.

The ARX structure plays a crucial role in symmetric-key primitives, with differential-linear (DL) attacks being among the most effective cryptanalysis techniques against ARX ciphers. In this paper, we present a systematic re-decomposition technique for DL distinguishers of ARX ciphers and identify for the first time the hourglass(-like) structural commonalities among optimal DL distinguishers searched out by various deduction techniques, also supported through comprehensive experiments, which motivate us to develop an efficient and generalized approach to construct optimal hourglass(-like) structural DL distinguishers. Our method yields significant advances when applied to \speck, \alzette, and the underlying permutations of \siphash{} and \chaskey: (1) the first 11- to 14-round DL distinguishers of \alzette; (2) the first (valid) DL distinguishers for 11-round \speck32, 12-round \speck48, and 16-round \speck96; (3) deterministic (correlation $\pm1$) 3-round DL distinguishers for \siphash-2-4 and significantly improved 4-round ones. All these distinguishers are equipped with both theoretical and experimental verifications. We further analyze ARX-based Latin dance stream ciphers, achieving improved DL distinguishers for 7/7.25-round \chacha, 8-round \salsa, and 5.5-round \forro. Though some of the improvements are not significant, we have verified the effectiveness of our method across a broader range of instances. This work provides new insights for DL distinguisher construction and enhances understanding of the security of ARX ciphers.

Quantum cryptanalysis of block ciphers with Grover’s search requires synthesis of round function, where the non-linear S-boxes dominate the circuit cost. Efficient quantum implementations of these S-boxes are a bottleneck for cryptanalysis. In this work, we address this problem and present new generic strategy for synthesis of quantum circuit for large S-boxes that reduces the NISQ-era transpiled depth after decomposition into the hardware-oriented universal basis gate set u+cx. We introduce two-phase MILP-based, ancilla-aware synthesis framework for large S-boxes. Phase 1 determines which monomials will be synthesised globally, and how they are reused across outputs. This reduces redundancy and avoids high-degree terms that would lead to deep ladders. Phase 2 arranges the selected monomials into parallel layers. Here the solver explicitly accounts for ancilla usage, balancing the trade-off between fewer layers (smaller depth) and larger ancilla. MILP based synthesis show decisive, multi-fold reductions in transpiled depth in universal basis gate set u+cx. For SKINNY and ZUC S0 S-boxes, our synthesis reduces transpiled depth by factors of 18 and 9, respectively, with ancilla usage raised only to 10 and 13 qubits. For higher-degree S-boxes such as AES, SM4, and ZUC S1, we achieve 5 times reduction in transpiled depth by trading additional ancillas, increasing the budget from 5 to 22. To our knowledge, this is the first demonstration of ancilla-aware, globally optimised synthesis of 8-bit cryptographic S-boxes. By aligning primitive synthesis with transpiled cost, our method establishes a new baseline for depth-optimised resource estimation in the quantum cryptanalysis of symmetric primitives.

Recently, Cotan and Teseleanu (NordSec 2023) published an RSA-like cryptosystem. While in RSA, the public exponent $e$ and the private exponent $d$ are related by the equation $ed - k(p-1)(q-1) = 1$, in their scheme, $e$ and $d$ are related to the equation $ed - k(p^n-1)(q^n-1) = 1$ for some positive integer $n$. Teseleanu (CSCML 2024) showed that one can factor the modulus $N$ using a lattice attack if $d$ is small. In this paper, we extend his attack by showing that if the private exponent is either too small or too large, one can factor $N$ in polynomial time by solving the generalized equation $eu - (p^n - 1)(q^n - 1)v = \pm 1$ using lattice reduction techniques.

As an ISO/IEC standard, RIPEMD-160 has been extensively studied for (Semi-Free-Start) collision attacks. A significant breakthrough was achieved at FSE 2024 with the first 41-, 42-, and 43-step SFS collision attacks, which leveraged an automatic search model (EUROCRYPT 2023) and a message modification strategy (FSE 2020). However, these attacks are limited by reliance on heuristic objective functions and suboptimal message modification techniques. This paper enhances the existing framework from two perspectives. Firstly, we refine the automatic search model by incorporating a holistic objective function that considers all critical probability components, moving beyond simple Hamming weight. Secondly, we introduce two generic techniques to further improve (SFS) collision attacks: the first application of differential clustering and a dedicated message modification strategy. As a result, we present the first valid SFS collision attack on 44-step RIPEMD-160. Additionally, we significantly reduce the time complexities of existing attacks on 41-, 42-, and 43-step variants, making it feasible to find colliding message pairs for 41- and 42-step versions within practical time for the first time.

Encrypted search focuses on protecting sensitive data in outsourced environments while enabling private queries. Although standard encrypted search algorithms are efficient, they often leak some information about the queries and data. One such leakage is the access pattern on the outsourced storage. Recent leakage-abuse attacks have exploited this seemingly harmless leakage to successfully recover both queries and data, shifting research priorities towards finding the right balance between privacy and performance. While some proposals leverage oblivious RAM or other oblivious data structures to hide the access pattern, they typically incur significant bandwidth costs. In response, researchers have developed new schemes that ensure access leakage satisfies differential privacy (DP). Yet the security implications of these new guarantees remain unclear. Especially, compared with conventional differential privacy, the application and threat model are significantly different. To understand these implications, we investigate two concrete instances of (encrypted) range-query schemes (appeared in SODA ’19 and CCS ’22) that achieve differentially private access. We analyze their security guarantees using inference attacks to recover queries and data on real-world datasets. Our findings raise a critical concern that ensuring access leakage is differentially private either falls short of providing strong security for the queries and data, diverging from the initial goals, or offers only weak security but at a high efficiency/correctness cost. As part of our analysis, we also propose a generic security definition for DP access, and identify two general techniques for leakage mitigation, bucketization and partitioning, that may be of independent interest.

The block cipher GOST 28147-89 was the Russian Federation encryption standard for over 20 years, and is still one of its two standard block ciphers. GOST is a 32-round Feistel construction, whose security benefits from the fact that the S-boxes used in the design are kept secret. In the last 10 years, several attacks on the full 32-round GOST were presented. However, they all assume that the S-boxes are known. When the S-boxes are secret, all published attacks either target a small number of rounds, or apply for small sets of weak keys. In this paper we present the first practical-time attack on GOST with secret S-boxes. The attack works in the related-key model and is faster than all previous attacks in this model which assume that the S-boxes are known. The complexity of the attack is less than $2^{27}$ encryptions. It was fully verified, and runs in a few seconds on a PC. The attack is based on a novel type of related-key differentials of GOST, inspired by local collisions. Our new technique may be applicable to certain GOST-based hash functions as well. To demonstrate this, we show how to find a collision on a Davies-Meyer construction based on GOST with an arbitrary initial value, in less than $2^{10}$ hash function evaluations.

Regular Syndrome Decoding (RSD) is a variant of the traditional Syndrome Decoding (SD) problem, where the error vector is divided into consecutive, equal-length blocks, each containing exactly one nonzero element. Recently, RSD has gained significant attention due to its extensive applications in cryptographic constructions, including MPC, ZK protocols, and more. The computational complexity of RSD has primarily been analyzed using two methods: Information Set Decoding (ISD) approach and algebraic approach. In this paper, we introduce a new hybrid algorithm for solving the RSD problem. This algorithm can be viewed as replacing the meet-in-the-middle enumeration in ISD with a process that solves quadratic equations. Our new algorithm demonstrates superior performance across a wide range of concrete parameters compared to previous methods, including both ISD and algebraic approaches, for parameter sets over both large fields (q = 2^128) and binary fields (q = 2). For parameter sets used in prior works, our algorithm reduces the concrete security of RSD by up to 20 bits compared to the state-of-the-art algorithms. We also provide an asymptotic analysis, identifying a broader parameter region where RSD is solvable in polynomial time compared to ISD and algebraic methods over binary fields. Additionally, we apply our algorithm to evaluate the security of the ZK protocol Wolverine (IEEE S&P 2021) and the OT protocol Ferret (ACM CCS 2020). Our results reduce the security level of Wolverine, which targets a 128-bit security level, to about 111 bits, and also marginally lowers the security of Ferret below the targeted 128-bit level for the first time.

The LowMC-based post-quantum signature scheme Picnic was selected as a third-round candidate for NIST PQC, attracting wide attention to the design of efficient and secure post-quantum signature schemes using Symmetric Techniques for Advanced Protocols (STAP). Symmetric primitives designed for advanced protocols such as secure multi-party computation (MPC), fully homomorphic encryption (FHE), and zero-knowledge (ZK) proof systems, with the goal of reducing the number of multiplication operations, are referred to as arithmetic-oriented (AO) primitives. These cryptographic primitives are typically constructed over large finite fields, which makes classical statistical analysis methods like differential and linear cryptanalysis inefficient. Due to their inherent algebraic properties, the mainstream security evaluation approaches are based on algebraic attacks. In this paper, we analyze the security of the MPC-friendly primitives \textsc{Rain} (CCS 2022) and AIM (CCS 2023) used in the post-quantum signature schemes Rainier and AIMer. Existing algebraic attacks on \textsc{Rain} and AIM were conducted over $\mathbb{F}_2$. We propose a novel algebraic attack over $\mathbb{F}_{2^n}$ that uses the polynomial decomposition to reduce degrees of equations. By further combining with the guess-and-determine technique, meet-in-the-middle modeling, and resultant, we are able to attack \textsc{Rain} and the full AIM. Specifically, we successfully attacked 2-round \textsc{Rain} with $2^{73.7}/2^{107.0}/2^{138.9}$ primitive calls, 3-round \textsc{Rain} with $2^{167.6}/2^{243.6}/2^{319.1}$ field operations, for the $128/192/256$-bit key. For the full AIM, we successfully attacked it with $2^{114.0}/2^{163.2}/2^{228.3}$ primitive calls for the $128/192/256$-bit key. The attack complexities mainly lie in solving univariate polynomial equations and computing resultants, and hence the complexity evaluations are accurate.

Constant-time implementations are a cornerstone of secure cryptographic systems, particularly in the context of key exchange protocols and digital signature schemes. These implementations are designed to eliminate timing side-channel vulnerabilities by ensuring that the program’s execution time is independent of secret data. A fundamental building block for achieving constant-time behavior is the conditional move operation. Unlike traditional branching constructs (such as if statements), which may introduce data-dependent timing variations, conditional moves allow developers to write logic that behaves identically at the hardware level regardless of input values. As a result, they are widely used in cryptographic libraries and standards to ensure both functional correctness and resistance to timing attacks. In this work, we describe our efforts to implement elliptic curve cryptography with some immunity against certain power leakage side-channel attacks, using standard C and Rust code.

At CRYPTO 2020, Liu et al. demonstrated that many differentials on Gimli are, in fact, incompatible. Similar incompatibilities also arise in related-key differentials on AES, which are typically addressed in an ad-hoc manner by incorporating additional constraints into the searching models. However, such ad-hoc methods are insufficient to eliminate all incompatibilities and may still produce false positive related-key differentials. At CRYPTO 2022, a novel approach was introduced that combines a Constraint Programming (CP) tool with a triangulation algorithm to search for rebound attacks against AES-like hashings. In this paper, we extend and unify these techniques to develop a comprehensive related-key differential search model. Our model not only identifies valid related-key differentials for AES and similar ciphers, but also enables immediate verification of the existence of at least one key pair satisfying the differentials. Using this enhanced automatic tool, we discover new related-key differentials for full-round AES-192, AES-256, Kiasu-BC, and for round-reduced Deoxys-BC. Based on these findings, we present full-round limited-birthday chosen-key distinguishing attacks on AES-192, AES-256, and Kiasu-BC, as well as the first chosen-key distinguisher on reduced-round Deoxys-BC. Furthermore, we identify, for the first time, a limited-birthday distinguisher on 9-round Kiasu-BC with practical complexities.

In its decisional form, the module-Lattice Isomorphism Problem (decision module-LIP) has received first attention in a paper by Ling, Liu and Mendelsohn. The authors gave a polynomial-time algorithm to distinguish spinor genera within the genus of a quadratic binary $\mathcal{O}_F$-lattice, assuming that $\mathcal{O}_F$ is a principal ideal domain. However, this algorithm would not impact cryptographic schemes based on decision module-LIP for lattices such as those employed in HAWK, i.e., for binary $\mathcal{O}_K$-lattices equipped with an Hermitian form (with $K$ a cyclotomic number field). Motivated by HAWK's framework, we investigate a concept that serves as an analogue of the spinor genus for Hermitian lattices, called special genus. This notion was studied by Shimura who provided a complete set of invariants for describing special genera. Building on this result, we propose an algorithm to determine whether two Hermitian lattices belong to the same special genus. Specifically for HAWK's lattice and siblings, our algorithm runs in classical polynomial-time. Nevertheless we provide numerical evidence suggesting that the ability to distinguish special genera does not, in practice, constitute a significative advantage for solving decision module-LIP.

Ethereum, a leading blockchain platform, relies on incentive mechanisms to improve its stability. Recently, several attacks targeting the incentive mechanisms have been proposed. Examples include the so-called reorganization attacks that cause blocks proposed by honest validators to be discarded to gain more rewards. Finding these attacks, however, heavily relies on expert knowledge and may involve substantial manual effort. We present BunnyFinder, a semi-automated framework for finding incentive flaws in Ethereum. BunnyFinder is inspired by failure injection, a technique commonly used in software testing for finding implementation vulnerabilities. Instead of finding implementation vulnerabilities, we aim to find design flaws. Our main technical contributions involve a carefully designed strategy generator that generates a large pool of attack instances, an automatic workflow that launches attacks and analyzes the results, and a workflow that integrates reinforcement learning to fine-tune the attack parameters and identify the most profitable attacks. We simulate a total of 9,354 attack instances using our framework and find the following results. First, our framework reproduces five known incentive attacks that were previously found manually. Second, we find three new attacks that can be identified as incentive flaws. Finally and surprisingly, one of our experiments also identified two implementation flaws.

The Hamming Quasi-Cyclic (HQC) key encapsulation mechanism (KEM), recently selected by NIST for standardization in the Post-Quantum Cryptography (PQC) process, distinguishes itself through its efficiency, robust design based on hard decoding problems in coding theory, and well-characterized decryption failure rates. Despite its selection, practical security concerns arise from implementation threats, particularly those exploiting plaintext-checking (PC) oracles. While multi-value PC (MV-PC) and full decryption (FD) oracle attacks have been extensively studied in the context of lattice-based cryptography, their applicability to code-based schemes like HQC has remained relatively unexplored. In this work, we present the first MV-PC and FD oracle attacks targeting code-based KEMs, specifically on HQC. Our MV-PC attack significantly reduces the required oracle queries compared to previous PC oracle-based methods and holds implications for side-channel, key-mismatch, and fault-injection attacks. Our FD attack exhibits remarkable efficiency in trace complexity, achieving secret key recovery for hqc-128 with just two queries to a perfect oracle, and four queries for hqc-192 and hqc-256. Simulations further demonstrate the robustness of our MV-PC and FD oracle attacks against imperfect oracle responses. We experimentally validate the new attacks on an ARM Cortex-M4 microcontroller, highlighting the critical need for robust countermeasures. In particular, on such a platform, substantial leakage during operations like syndrome computation poses significant challenges to the efficiency of masking techniques in mitigating FD oracle attacks.

MIFARE Classic smart cards, developed and licensed by NXP, are widely used but have been subjected to numerous attacks over the years. Despite the introduction of new versions, these cards have remained vulnerable, even in card-only scenarios. In 2020, the FM11RF08S, a new variant of MIFARE Classic, was released by the leading Chinese manufacturer of unlicensed "MIFARE compatible" chips. This variant features specific countermeasures designed to thwart all known card-only attacks and is gradually gaining market share worldwide. In this paper, we present several attacks and unexpected findings regarding the FM11RF08S. Through empirical research, we discovered a hardware backdoor and successfully cracked its key. This backdoor enables any entity with knowledge of it to compromise all user-defined keys on these cards without prior knowledge, simply by accessing the card for a few minutes. Additionally, our investigation into older cards uncovered another hardware backdoor key that was common to several manufacturers.

QARMAv2 represents a family of lightweight block ciphers introduced in ToSC 2023. This new iteration, QARMAv2, is an evolution of the original QARMA design, specifically constructed to accommodate more extended tweak values while simultaneously enhancing security measures. This family of ciphers is available in two distinct versions, referred to as QARMAv2-$b$-$s$, where ‘$b$’ signifies the block length, with options for both 64-bit and 128-bit blocks, and ‘$c$’ signifies the key length. In this paper, for the first time, we present differential fault analysis (DFA) of all the QARMAv2 variants- QARMAv2-64, and QARMAv2-128 by introducing an approach to utilize the fault propagation patterns at the nibble level, with the goal of identifying relevant faulty ciphertexts and vulnerable fault positions. This technique highlights a substantial security risk for the practical implementation of QARMAv2. By strategically introducing six random nibble faults into the input of the $(r − 1)$-th and $(r − 2)$-th backward rounds within the $r$-round QARMAv2-64, our attack achieves a significant reduction in the secret key space, diminishing it from the expansive $2^{128}$ to a significantly more smaller set of size $2^{32}$. Additionally, when targeting QARMAv2-128-128, it demands the introduction of six random nibble faults to effectively reduce the secret key space from $2^{128}$ to a remarkably reduced $2^{24}$. To conclude, we also explore the potential extension of our methods to conduct DFA on various other iterations and adaptations of the QARMAv2 cryptographic scheme. To the best of our knowledge, this marks the first instance of a differential fault attack targeting the QARMAv2 tweakable block cipher family, signifying an important direction in cryptographic analysis.

AES has cemented its position as the primary symmetric-key primitive for a wide range of cryptographic applications, which motivates the analysis on the concrete security of AES in practical instantiations, for instance, the collision resistance of AES-based hashing, the key commitment security of AES-based authenticated encryption schemes, and the one-wayness of AES-based one-way functions in MPC/ZK protocols. In this work, we further advance the meet-in-the-middle (MITM) attack framework on AES-like constructions. We introduce single-color initial structure (SCIS), which leverages new structural insights to reduce the complexity of neutral word generation, a critical bottleneck in MITM attacks. As a result, we yield a series of improved results on AES over the state-of-the-art, including the first classical one-block collision attack on 7-round AES-MMO/MP, marking the first round advancement in over a decade and matching the best attack round in the quantum setting, as well as the first one-block collision attack on 4-round AES-128-DM, bridging the gap highlighted by Taiyama et al. at Asiacrypt 2024 from a non-differential-based approach. Additionally, we provide a comprehensive list of new results on the security margins of AES-192, AES-256, Rijndael-192, and Rijndael-256 in multiple attack settings.

Index Calculus (IC) algorithm is the most effective probabilistic algorithm for solving discrete logarithms over finite fields of prime numbers, and it has been widely applied to cryptosystems based on elliptic curves. Since the IC algorithm was proposed in 1920, the research on it has never stopped, especially discretization of prime numbers on the finite fields, both the algorithm itself and its application have been greatly developed. Of course, there has been some research on elliptic curves,but with little success. For the IC algorithm, scholars pay more attention to how to improve the probability of solving and reduce the time complexity of calculation. It is the first time for the IICA to study the optimization problem of the IC by using the method of integer. However, the IICA only studies the case of integer up, and fails to consider the case of integer down. It is found that the integer direction of the IICA can be integer up or integer down, but the concept of modular multiplication needs to be used when integer down. After optimizing the IICA, the probability of successful solution of discrete logarithm is increased by nearly 2 times, and the number of transformations is also reduced to a certain extent, thus reducing the time complexity of solution. The re-optimized the IC algorithm greatly improves the probability of successful the IC solution. This research result poses a serious challenge to cryptosystems based on finite fields of prime numbers.

NIST has recently selected CRYSTALS-Kyber as a new public key encryption and key establishment algorithm to be standardized. This makes it important to evaluate the resistance of CRYSTALS-Kyber implementations to side-channel attacks. Software implementations of CRYSTALS-Kyber have already been thoroughly analysed. The discovered vulnerabilities helped improve the subsequently released versions and promoted stronger countermeasures against side-channel attacks. In this paper, we present the first attack on a protected hardware implementation of CRYSTALS-Kyber. We demonstrate a practical message (shared key) recovery attack on the first-order masked FPGA implementation of Kyber-512 by Kamucheka et al. (2022) using power analysis based on the Hamming distance leakage model. The presented attack exploits a vulnerability located in the masked message decoding procedure which is called during the decryption step of the decapsulation. The message recovery is performed using a profiled deep learning-based method which extracts the message directly, without extracting each share explicitly. By repeating the same decapsulation process multiple times, it is possible to increase the success rate of full shared key recovery to 99%.

Isogeny-based cryptosystems continue to show promise in post-quantum cryptography. In recent years, numerous constructions have been proposed, one of which is POKÉ, a compact and efficient public-key exchange system that uses higher-dimensional isogenies. This paper leverages a well-known adaptive attack on SIDH by Galbrath, Petit, Shani and Ti, and demonstrates a similar attack on POKÉ, when given a key exchange oracle with the same assumptions as those posed by Galbraith et al. This attack relies on the user to employ long-term static keys which is against the intent of the designers of POKÉ. Indeed, this attack provides further evidence that POKÉ should not be used with a long-term static key.

Perceptual hash functions have been designed to detect multimedia copyright violations and illegal content. To achieve their purpose, they have to maps inputs that are perceived as similar to close outputs. For several widely used designs the design strategy or even the design details are proprietary. Governments consider extending these functions to Client-Side Scanning (CSS) for end-to-end encrypted services, verifying content against illegal material before encryption. In 2021, Apple presented a detailed proposal for CSS based on the NeuralHash perceptual hash function. After strong criticism pointing out privacy and security concerns, Apple withdrew the proposal, but NeuralHash remains deployed on all devices, with its current purpose undisclosed. Brute-force collisions for NeuralHash (with a 96-bit result) require $2^{48}$ evaluations. Shortly after NeuralHash’s release, researchers showed it is easy to craft perceptually dissimilar collisions, enabling false incrimination in CSS by sending an innocent image with the same hash as illegal content. This work shows a more serious weakness: when inputs are restricted to a set of human faces, random collisions are highly likely to occur in input sets of size $2^{16}$. Unlike the targeted attack, our black-box attacks require no knowledge of the hash function’s design. We also show a high false negative rate (pictures that should share the same hash but do not). We show the generality of our approach by attacking PhotoDNA, Microsoft’s widely deployed 1152-bit perceptual hash. Small input sets yield near-collisions after only $2^{14.6}$ or $2^{17}$ operations, depending on the threshold. These results show that current perceptual hash designs are unsuitable for large-scale client scanning, producing high false positive and false negative rates, and highlight the need to reassess their security and feasibility, particularly for large-scale applications where privacy risks and false positives have serious consequences.

We utilise the Maiorana-McFarland secondary construction to generate bent functions. Our process begins with a bent function of dimension $s-r$ and results in a new function of dimension $s+r$. To achieve this, we first consider affine spaces from which we can derive bent functions. Furthermore, we derive an expression to obtain bent functions by applying the case $r=1$ multiple times. We demonstrate that the two resulting expressions are distinct. In nearly all constructions, the bent functions we obtain are balanced when the domain is restricted to vectors with an even Hamming weight. Finally, we present an example algorithm, focusing on the case where $r=2$, as well as another algorithm that employs $r=1$ twice.

Recent work showed that ML-based attacks on Learning with Errors (LWE), a hard problem used in post-quantum cryptography, outperform classical algebraic attacks in certain settings. Although promising, ML attacks struggle to scale to more complex LWE settings. Prior work connected this issue to the difficulty of training ML models to do modular arithmetic, a core feature of the LWE problem. To address this, we develop techniques that significantly boost the performance of ML models on modular arithmetic tasks—enabling the models to sum up to $N=128$ elements modulo $q \le 974269$. Our core innovation is the use of custom training data distributions and a carefully designed loss function that better represents the problem structure. We apply an initial proof of concept of our techniques to LWE specifically and find that they allow recovery of 2x harder secrets than prior work. Our techniques also help ML models learn other well-studied problems better, including copy, associative recall, and parity, motivating further study.

Information Set Decoding (ISD) refers to a class of algorithms designed to decode arbitrary linear codes over finite fields. It is among the state-of-the-art methods for solving the Syndrome Decoding Problem (SDP), which lies at the core of code-based cryptography. Since most cryptographic systems operate over the binary field, ISD algorithms are generally designed for this setting. However, emerging cryptographic systems, such as SDitH, that rely on the SDP over non-binary fields, make research into non-binary ISD increasingly relevant. While there have been numerous improvements to ISD algorithms for binary fields, generalizations of these improvements to non-binary fields have resulted in only modest runtime improvements. The only technique which applies specifically to non-binary fields was proposed only last year by Carrier, Hatey and Tillich and consists in solving the SDP over the projective space. In this paper we continue along this line of research and introduce a novel technique to perform ISD in non-binary fields. Our key idea consists in speeding-up the enumeration of low weight vectors by enumerating only the supports and not the actual values, since these can be recovered by solving a (small) linear system. This idea is at the core of LA-ISD, the first algorithm we propose in the paper. We further enhance it by exploiting a meet-in-the-middle search leading to MitM-LA, the second algorithm we propose in this paper. We analyze our algorithms in both the finite and asymptotic regimes and show that they compare well with competitive solutions. In particular, LA-ISD results in the best memory-less algorithm, while MitM-LA is (slightly) faster than the state-of-the-art algorithm by Carrier, Hatey and Tillich for many parameter-sets as well as asymptotically.

FRAST is a TFHE-friendly stream cipher that was published at FSE 2025. The cipher is defined over $\mathbb{Z}_{16}$, and makes extensive use of negacyclic S-boxes over $\mathbb Z_{16}$ as they are less costly in TFHE. Like many FHE-friendly ciphers, FRAST randomizes some of its components to increase its security against statistical attacks. In the case of FRAST, some S-boxes are randomized using an XOF that takes a nonce as input. In this work, we point out a strong structural property of the full FRAST permutation, which leads to a much simpler alternative representation of the primitive. We study the consequences of this representation and find a weak key space of non-negligible size (i.e., much larger than $2^{128}$) on which every ciphertext leaks one bit of plaintext. This corresponds to a distinguishing attack on the full FRAST in the weak-key setting. In particular, we emphasize that, apart from the structural property, the usage of negacyclic S-boxes further leads to a much larger weak-key space for our attack. Finally, we provide a general framework to mount a linear attack on FRAST in the average key setting. We briefly describe our approach in the end of the paper, and observe that standard assumptions expected to work in the context of linear cryptanalysis do not hold in the case of FRAST: our experiment indicate that a linear attack in the average key setting does not work as expected.

This paper presents an enhancement to cube-attack-like cryptanalysis by minimizing output-bit dependency on related key bits, thereby improving attack complexity. We construct two distinct initial states differing exclusively in predetermined bit positions. Through independent cube summation and state difference analysis, we observed reduced related key bits dependency for specific output bits. We validate our approach by targeting three Keccak keyed variants Ketje Minor, Keccak-MAC-512 and Keccak-MAC-384, developing a dedicated tool to recover all output-bit superpolies. Using our computational resources, we successfully attacked 4-round of Ketje Minor and 5-round of other variants, confirming both the method's validity and practical applicability. While the best known attacks on these structures reach 7-round, our results improve upon the 5-round. We construct our initial state configurations based on the automated method proposed by Bi et al.\ in \textit{Design, Codes and Cryptography} (2019), and compare our results with theirs. For the 4-round Ketje Minor, we reduce the time complexity from \(2^{20}\) to \(2^{16.8}\); for 5 round Keccak-MAC-512, from \(2^{34}\) to \(2^{31.2}\); and for 5 round Keccak-MAC-384, from \(2^{27.6}\) to \(2^{26.5}\).

Motivated by LowMC cryptanalysis challenge, research in recent years focuses more on attacking LowMC in \PICNIC application setting, \ie an attacker can see only a single plaintext/ciphertext pair. It can be noted that in the security proof of \PICNIC, LowMC is required to be secure under two plaintexts, it is thus meaningful to investigate the security of LowMC in this direction. Pioneered by Liu, Isobe and Meier at Crypto 2021, they combined algebraic techniques with difference enumeration attack, which could attack all three 4-round LowMC instances adopting full S-Box layers in \PICNIC with only two chosen plaintexts. However, the research on cryptanalysis of LowMC using two plaintext/ciphertext pairs is yet far from complete. Previous works using a single known plaintext are better than those with two plaintexts in terms of attack complexity or attacked rounds when considering a comparable success probability. In this paper, to address such counter-intuitive gaps between existing attacks on LowMC with full S-box layers using a single and two plaintext/ciphertext pairs, we first develop an algebraic key-derived attack framework, where an algebraic property of the key-derived difference is utilized to build an equation system with lower algebraic degree. This directly contributes to less cost for solving equation system and naturally works under known-plaintext setting, which can be further enhanced with chosen-plaintext attack setting. We then present an improved difference enumeration attack framework. Instead of enumerating all possible differences in the second round, variables for part of S-boxes in the second and third rounds are introduced to derive cubic equations, which will lead to fewer variables for the last round. Finally, applying our new attack frameworks to LowMC, we propose \text{8-round} attacks on LowMC for the very first time, which remain under known-plaintext setting. Moreover, we give the first attacks on three LowMC instances, \ie 129-bit block size of 6 rounds and 129-/192-bit block size of 7 rounds, which cannot be obtained using previous attacking methods. Also, previous attacks on LowMC from 4 to 7 rounds could be improved for almost all three LowMC instances in this paper. All these results, we believe, could be a positive answer that given one more pair, more information indeed can be gained to improve attacks on LowMC when compared to those using only a single plaintext. As well as our newly proposed algebraic key-derived attack framework, we hope that, could provide more insights into the cryptanalysis of LowMC with low allowable data complexity.

\xoodyak is a finalist of the NIST lightweight cryptography competition, offering both keyed and hash modes. After several years of cryptanalysis, the largest number of \xoodyak hash rounds for which actual collisions was still in vacancy. To the best of our knowledge, one of the most powerful collision attacks on hash functions based on sponge construction is the differential-based attacks using the S-box linearization technique proposed by Qiao \etal (EUROCRYPT 2017). However, the linearization technique requires a large number of degrees of freedom, making it challenging to apply to \xoodyak with a small outer part. On the other hand, the constraint-input and constraint-output imposed on the differential trail of \xoodoo permutation make the exhaustive search for relatively high-probability differential trails in collision attacks extremely costly. In this paper, we present critical observations regarding \xoodoo round function, particularly focusing on its unique $\theta$ and $\chi$ operation. These properties can be leveraged to manually design specific differential trails, referred to as \textit{loop} trails. To efficiently find practical collisions for up to 3 rounds, we develop a SAT model based on these \textit{loop} trails. Finally, we present the first practical collision on 2 rounds and a practical semi-free-start collision on 3 rounds of \xoodyak hash mode. Besides, we improve Dong \etal's (CRYPTO 2024) collision attack on 3-round \xoodyak-\hash from $2^{125.23}$ to $2^{100.93}$ using several linearization strategies. Since we focus on the analysis on collisions during the message absorbing phase of the hash modes, our results are applicable to both \xoodyak-\hash and \xoodyak-\xof.

The development of cryptographic schemes which remain secure in the post-quantum era is an urgent challenge, particularly in light of the growing ubiquity of low-power devices and the looming threat of quantum computing. Identity-Based Encryption (IBE) offers a compelling alternative to traditional Public Key Infrastructures by simplifying key management, but most classical IBE schemes rely on number-theoretic assumptions that are vulnerable to quantum attacks. In response, Koshiba and Takashima proposed a novel approach based on Isogenous Pairing Groups (IPGs) [ 12 ], claiming partial quantum resistance. In this work, we critically examine their construction and security claims and investigate upon the security in standard security notions. We show that the proposed scheme, despite its theoretical elegance, reduces to the Elliptic Curve Dis- crete Logarithm Problem (ECDLP) on supersingular curves, which can be broken in polynomial time by quantum algorithms in the IND-ID-CPA setting and in subexponential time classically in the authors security models. Our analysis reveals structural weaknesses inherent to the IPG framework, such as the use of explicit group elements in prime-order groups and exploitable isogeny homomorphisms, which undermine the post-quantum security of isogeny based cryptography. These findings suggest that IPG-based constructions, which use more than a single point per pairing group, are unlikely to provide robust post-quantum security. In Koshiba and Takashima’s proposed IBE system, we can directly observe this, when paying regards to not only the IBE’s Master Secret Key but also Users’ Secret Keys.

The Q-problem has been introduced as a new post-quantum hard problem for the development of new post-quantum cryptography. We present two man-in-the-middle and four key recovery attacks against the key exchange protocol based on the Q-problem. The man-in-the-middle attacks take negligible time and allow the attacker to recover the exchanged key. The most effective key recovery attack is computationally trivial and, unlike the other attacks presented in this work, cannot be prevented by countermeasures.

Kleptography was first proposed by Adam Young and Moti Yung in 1996, while algorithm substitution attack was introduced by Mi- hir Bellare et al. as a variation of kleptography in 2014 after the Dual EC incident with the confidential documents revelation by Edward Snowden. These two paradigms share a common goal: to enable attackers to embed covert capabilities into cryptographic implementations while maintaining the appearance of normal functionality. The goal of this paper is to con- solidate existing research on kleptographic attacks, integrate it into a uni- fied definition, and explore future directions for the research in this field. This paper begins by introducing and comparing the two major branches of kleptographic attacks: traditional kleptography and post-Snowden al- gorithm substitution attack, highlighting their theoretical distinctions, threat models, and historical development. Then, it analyzes the spe- cific goals that attackers aim to achieve through such subversions and propose a generalized definition of algorithm substitution attack that in- clude all the goals. The paper also presents practical examples framed within my definition and classify prior research works as either strong or weak attacks based on their structure and undetectability. Finally, it discusses the current landscape of research in kleptographic attacks, and then suggest future directions for the attack and defense perspectives.

The Regular Syndrome Decoding (RSD) problem, introduced nearly two decades ago, is a regular version of the Syndrome Decoding (SD) problem, where the noise vector is divided into consecutive, equally sized blocks, each containing exactly one noisy coordinate. Recently, RSD has gained renewed attention for its applications in Pseudorandom Correlation Generator (PCG) and more. Very recently, several works presented the improved algebraic approach (AGB for short) and ISD approach (including regular-ISD and regular-RP) to utilize the regular structure of noise vectors and provide a more precise security evaluation for the RSD problem. In this paper, we refine the AGB algorithm from a one-round process to a two-round process, and refer to the new algebraic algorithm as AGB 2.0. For each round, we guess a few noise-free positions, followed by a tailored partial XL algorithm. This interleaving strategy increases the probability of success by reducing the number of guessing noise-free positions and effectively lowers the problem's dimension in each round. By fine-tuning position guesses in each round and optimizing the aggregate running time, our AGB 2.0 algorithm reduces the concrete security of the RSD problem by up to $6$ bits for the parameter sets used in prior works, compared to the best-known attack. In particular, for a specific parameter set in Wolverine, the RSD security is $7$ bits below the $128$-bit target. We analyze the asymptotic complexity of algebraic attacks on the RSD problem over a finite field $\mathbb{F}$ with the field size $|\mathbb{F}|>2$, when the noise rate $\rho$ and code rate $R$ satisfy $\rho + R < 1$. If $n \rho R^2 = O(1)$ where $n$ is the noise length, the RSD problem over $\mathbb{F}$ can be solved in polynomial time, but it does not hold for the SD problem. We show that the ISD and its variants, including regular-ISD and regular-RP, are asymptotically less efficient than AGB for solving RSD problems with $R = o(1/\log(n))$ and $|\mathbb{F}| > e^{n\rho R}$.

Distributed social networking services (SNSs) recently received significant attention as an alternative to traditional, centralized SNSs, which have inherent limitations on user privacy and freedom. We provide the first in-depth security analysis of Nostr, an open-source, distributed SNS protocol developed in 2019 with more than 1.1 million registered users. We investigate the specification of Nostr and the client implementations and present a number of practical attacks allowing forgeries on various objects, such as encrypted direct messages (DMs), by a malicious user or a malicious server. Even more, we show a confidentiality attack against encrypted DMs by a malicious user exploiting a flaw in the link preview mechanism and the CBC malleability. Our attacks are due to cryptographic flaws in the protocol specification and client implementation, some of which in combination elevate the forgery attack to a violation of confidentiality. We verify the practicality of our attacks via Proof-of-Concept implementations and discuss how to mitigate them.

Multivariate cryptography is one of the challenging candidates for post-quantum cryptography. There exists a huge variety of proposals, most of them have been broken substantially. Multivariate schemes are usually constructed by applying two secret affine invertible transformations $\mathcal S,\mathcal T$ to a set of multivariate polynomials $\mathcal{F}$ (often quadratic). The secret polynomials $\mathcal{F}$ possess a trapdoor that allows the legitimate user to find a solution of the corresponding system, while the public polynomials $\mathcal G=\mathcal S\circ\mathcal F\circ\mathcal T$ look like random polynomials. In [Calderini, M., Caminata, A., Villa, I. A New Multivariate Primitive from CCZ Equivalence. J. Cryptol. 38, 25 (2025)], the authors addressed the above challenge by presenting a promising new way of constructing a multivariate scheme by considering the CCZ equivalence, which has been introduced and studied in the context of vectorial Boolean functions. The resulting proposal is called Pesto with security parameters $s,t,m,n,q$, where $n$ is the number of variables, $s,t,m\leq n$ and $q$ the size of the finite base field $\mathbb{F}_q$. In this paper we present an attack against Pesto by constructing an equivalent secret key from the public key. This attack has a precomputation phase with a complexity of \[\textrm{max}\left\{{\mathcal{O}\left(n^{6}(m-t)\right),\mathcal{O}\left(\frac{(m-t)(n-t)^2(n-t-s)q^s}{\mathcal{P}(q,n-t-s)}\right)}\right\}\] base field operations on average and an online complexity of \[\mathcal{O}(q^s(m-t)(n-t-s) \cdot \min(m-t,n-t-s) + q^st (n-t)^2 + q^st^3)\] base field operations to decipher a message or forge a signature, where $\mathcal{P}(q,k) := \prod_{i=1}^k (1-1/q^i)$. Thus, our attack breaks Pesto for any practical choice of the security parameters $n,m,s,t,q$ and renders the concrete construction underlying Pesto insecure.

In Neural Cryptanalysis, a deep neural network is trained as a cryptographic distinguisher between pairs of ciphertexts $(F(X), F(X \oplus \delta))$, where $F$ is either a random permutation or a block cipher, $\delta$ is a fixed difference. The AutoND framework aims to se neural distinguishers that are treated as a generic tool and discourages cipher-specific optimizations. On the other hand, works such as $[\text{LLS}^+24]$ obtain superior distinguishers by adding dedicated features, such as selected parts of the difference in the previous rounds, to the input of the neural distinguishers. In this paper, we study $\text{Generic Partial Decryption}$ as a feature engineering technique and integrate it within a fully automated pipeline, where we evaluate its effect independently of the number of pairs per sample, with which feature engineering is often combined. We show that this technique matches state-of-the-art dedicated approaches on Simon and Simeck. Additionally, we apply it to Aradi, and present a practical neural-assisted key recovery for 5 rounds, as well as a 7-rounds key recovery with $2^{70}$ time complexity. Additionally, we derive useful information from the neural distinguishers and propose a non-neural version of our 5-round key recovery.

Physical side-channel analysis (SCA) operates on the foundational assumption of access to known plaintext or ciphertext. However, this assumption can be easily invalidated in various scenarios, ranging from common encryption modes like Cipher Block Chaining (CBC) to complex hardware implementations, where such data may be inaccessible. Blind SCA addresses this challenge by operating without the knowledge of plaintext or ciphertext. Unfortunately, prior such approaches have shown limited success in practical settings. In this paper, we introduce the Deep Learning-based Blind Side-channel Analysis (DL-BSCA) framework, which leverages deep neural networks to recover secret keys in blind SCA settings. In addition, we propose a novel labeling method, Multi-point Cluster-based (MC) labeling, accounting for dependencies between leakage variables by exploiting multiple sample points for each variable, improving the accuracy of trace labeling. We validate our approach across four datasets, including symmetric key algorithms (AES and Ascon) and a post-quantum cryptography algorithm, Kyber, with platforms ranging from high-leakage 8-bit AVR XMEGA to noisy 32-bit ARM STM32F4. Notably, previous methods failed to recover the key on the same datasets. Furthermore, we demonstrate the first successful blind SCA on a desynchronization countermeasure enabled by DL-BSCA and MC labeling. All experiments are validated with real-world SCA measurements, highlighting the practicality and effectiveness of our approach.

Side-Channel Analysis (SCA) exploits physical vulnerabilities in systems to reveal secret keys. With the rise of Internet-of-Things, evaluating SCA attacks has become crucial. Profiling attacks, enhanced by Deep Learning-based Side-Channel Analysis (DLSCA), have shown significant improvements over classical techniques. Recent works demonstrate that ensemble methods outperform single neural networks. However, almost every existing ensemble selection method in SCA only picks the top few best-performing neural networks for the ensemble, which we coined as Greedily-Selected Method (GSM), which may not be optimal. This work proposes Evolutionary Avenger Initiative (EAI), a genetic algorithm-driven ensemble selection algorithm, to create effective ensembles for DLSCA. We investigate two fitness functions and evaluate EAI across four datasets, including \AES and \ascon implementations. We show that EAI outperforms GSM, recovering secrets with the least number of traces. Notably, EAI successfully recovers secret keys for \ascon datasets where GSM fails, demonstrating its effectiveness.

With the advancement of deep learning techniques, Deep Learning-based Non-profiled Side-Channel Analysis (DL-NSCA) can automatically learn and combine features, making it a promising method that can skip the manual and precise selection of Points of Interest (PoIs). Existing DL-NSCA methods assume that the attacker can identify a short leakage interval (usually less than 5000 points) containing PoIs from raw traces (more than 100,000 points) and then feed the leakage interval into the neural network to recover the key. However, in practice, the attacker often faces a black-box scenario with unknown underlying implementations, making locating the short interval from raw traces challenging, especially when masking countermeasures exist. To address this issue, we propose a lightweight end-to-end DL-NSCA model called convWIN-MCR, which consists of a performance-optimizing component, convWIN, and an accelerator component, MCR. It can efficiently process raw traces without the need to manually identify the short leakage interval. On the public dataset ASCADv1, while the state-of-the-art model Multi-Output Regression (MOR) requires 28,000 traces and 24 minutes to recover the key from the leakage interval with 1,400 feature points, our framework only requires 6,000 traces in 13 minutes to directly analyze raw traces with 250,000 feature points. To further validate the practical applicability of our framework, we successfully crack a commercial USIM card by analyzing its raw traces and recovering its 128-bit AES key.

In this paper, we present a simple meet-in-the-middle attack that requires low data and memory resources. To evaluate the complexity of the attack, we also propose an automated tool that calculates the time, data, and memory complexities based on the suggested matching points. Our method operates at the bit level and employs a known-plaintext attack, with no constraints on the attacker's choice of data. We apply our tool on various lightweight block ciphers, including CRAFT, Midori, WARP, PRESENT, and ARADI. For CRAFT, our tool successfully identified an attack targeting 15 rounds using 3 known plaintexts. In the case of Midori64 and Midori128, the tool proposed attacks on 5 rounds with 16 known plaintexts and 7 rounds with 3 known plaintexts, respectively. For WARP, the tool discovered an attack on 18 rounds utilizing 7 known plaintexts. Additionally, for PRESENT80, the tool identified an attack on 6 rounds with 18 known plaintexts, and for ARADI, an attack on 5 rounds with 28 known plaintexts was determined.

Cascader has been introduced as a new key exchange protocol based on iterative multiplicative recurrence. This short note presents a practical shared key recovery attack on the Cascader key exchange protocol. This note also shows that Cascader as a hash function is not collision resistant, presents a new upper bound on the output space of Cascader and shows that a Cascader-based KDF is not secure against an Adaptive Chosen Public Inputs Attack (CPM).