Fully Homomorphic Encryption (FHE) is a cryptographic primitive that allows performing arbitrary operations on encrypted data. Since the conception of the idea in [RAD78], it was considered a holy grail of cryptography. After the first construction in 2009 [Gen09], it has evolved to become a practical primitive with strong security guarantees. Most modern constructions are based on well-known lattice problems such as Learning with Errors (LWE). Besides its academic appeal, in recent years FHE has also attracted significant attention from industry, thanks to its applicability to a considerable number of real-world use-cases. An upcoming standardization effort by ISO/IEC aims to support the wider adoption of these techniques. However, one of the main challenges that standards bodies, developers, and end users usually encounter is establishing parameters. This is particularly hard in the case of FHE because the parameters are not only related to the security level of the system, but also to the type of operations that the system is able to handle. In this paper, we provide examples of parameter sets for LWE targeting particular security levels that can be used in the context of FHE constructions. We also give examples of complete FHE parameter sets, including the parameters relevant for correctness and performance, alongside those relevant for security. As an additional contribution, we survey the parameter selection support offered in open-source FHE libraries.

Side-channel analysis (SCA) does not aim at the algorithm's weaknesses but rather its implementations. The rise of machine learning (ML) and deep learning (DL) is giving adversaries advanced capabilities to perform stealthy attacks. In this paper, we propose DL-SCADS, a DL-based approach along with signal decomposition techniques to leverage the power of secret key extraction from post-silicon EM/power side-channel traces. We integrate previously proven effective ideas of model ensembling and automated hyperparameter tuning with signal decomposition to develop an efficient and robust side-channel attack. Extensive experiments are performed on Advanced Encryption Standard (AES) and Post-Quantum Cryptography (PQC) to demonstrate the efficacy of our approach. The evaluation of the performance of the side-channel attack employing various decomposition techniques and comparison with the proposed approach in a range of datasets are also tabulated.

The RSA (Rivest-Shamir-Adleman) cryptosystem is a fundamental algorithm of public key cryptography and is widely used across various information domains. For an RSA modulus represented as $N = pq$, with its factorization remaining unknown, security vulnerabilities arise when attackers exploit the key equation $ed-k(p-1)(q-1)=1$. To enhance the security, Murru and Saettone introduced cubic Pell RSA --- a variant of RSA based on the cubic Pell equation, where the key equation becomes $ed-k(p^2+p+1)(q^2+q+1)=1$. In this paper, we further investigate the security implications surrounding the generalized key equation $eu-(p^2+p+1)(q^2+q+1)v=w$. We present a novel attack strategy aimed at recovering the prime factors $p$ and $q$ under specific conditions satisfied by $u$, $v$, and $w$. Our generalized attack employs lattice-based Coppersmith's techniques and extends several previous attack scenarios, thus deepening the understanding of mathematical cryptanalysis.

This paper investigates the Mersenne number-based $\mathsf{AJPS}$ cryptosystem, with a particular focus on its associated hard problem. Specifically, we aim to enhance the existing lattice-based attack on the Mersenne low Hamming ratio search problem. Unlike the previous approach of directly employing lattice reduction algorithm, we apply the lattice-based method to solving polynomial equations derived from the above problem. We extend the search range for vulnerabilities in weak keys and increase the success probability of key recovery attack. To validate the efficacy and accuracy of our proposed improvements, we conduct numerical computer experiments. These experiments serve as a concrete validation of the practicality and effectiveness of our improved attack.

Side-channel attacks exploit information leaked through non-primary channels, such as power consumption, electromagnetic emissions, or timing, to extract sensitive data from cryptographic devices. Over the past three decades, side-channel analysis has evolved into a mature research field with well-established methodologies for analyzing standard cryptographic algorithms like the Advanced Encryption Standard (AES). However, the integration of side-channel analysis with formal methods remains relatively unexplored. In this paper, we present a hybrid attack on AES that combines side-channel analysis with SAT. We model AES as a SAT problem and leverage hints of the input and output values of the S-boxes, extracted via profiled deep learning-based power analysis, to solve it. Experimental results on an ATXmega128D4 MCU implementation of AES-128 demonstrate that the SAT-assisted approach consistently recovers the full encryption key from a single trace, captured from devices different from those used for profiling, within one hour. In contrast, without SAT assistance, the success rate remains below 80% after 26 hours of key enumeration.

Sieving using near-neighbor search techniques is a well-known method in lattice-based cryptanalysis, yielding the current best runtime for the shortest vector problem in both the classical [BDGL16] and quantum [BCSS23] setting. Recently, sieving has also become an important tool in code-based cryptanalysis. Specifically, using a sieving subroutine, [GJN23, DEEK24] presented a variant of the information-set decoding (ISD) framework, which is commonly used for attacking cryptographically relevant instances of the decoding problem. The resulting sieving-based ISD framework yields complexities close to the best-performing classical algorithms for the decoding problem such as [BJMM12, BM18]. It is therefore natural to ask how well quantum versions perform. In this work, we introduce the first quantum algorithms for code sieving by designing quantum variants of the aforementioned sieving subroutine. In particular, using quantum-walk techniques, we provide a speed-up over the best known classical algorithm from [DEEK24] and over a variant using Grover's algorithm [Gro96]. Our quantum-walk algorithm exploits the structure of the underlying search problem by adding a layer of locality-sensitive filtering, inspired by the quantum-walk algorithm for lattice sieving from [CL21]. We complement our asymptotic analysis of the quantum algorithms with numerical results, and observe that our quantum speed-ups for code sieving behave similarly as those observed in lattice sieving. In addition, we show that a natural quantum analog of the sieving-based ISD framework does not provide any speed-up over the first presented quantum ISD algorithm [Ber10]. Our analysis highlights that the framework should be adapted in order to outperform the state-of-the-art of quantum ISD algorithms [KT17, Kir18].

Arithmetization-Oriented (AO) cryptographic algorithms operate on large finite fields. The most threatening attack on such designs is the Gröbner basis attack, which solves the equation system encoded from the cryptanalysis problem. However, encoding a primitive as a system of equations is not unique, and finding the optimal one with low solving complexity is a challenge. This paper introduces an automatic tool that converts the CICO problem into a Mixed-Integer Quadratic Constraint Programming (MIQCP) model, using integer variables and constraints to track degree propagation and determine variable introduction points. The optimal MIQCP solution provides the lowest solving complexity. We build models for Griffin, Anemoi, and Ciminion permutations to cover modules comprehensively. Experiments show reduced Gröbner basis attack complexity, lower than designers’ bounds for small numbers of rounds, e.g. up to 8 rounds for Griffin.This tool can be used for security evaluation against Gröbner basis attack in new designs.

Several key exchange protocols based on tropical circulant matrices were proposed in the last two years. In this paper, we show that protocols offered by M. Durcheva [M. I. Durcheva. TrES: Tropical Encryption Scheme Based on Double Key Exchange. In: Eur. J. Inf. Tech. Comp. Sci. 2.4 (2022), pp. 11–17], by B. Amutha and R. Perumal [B. Amutha and R. Perumal. Public key exchange protocols based on tropical lower circulant and anti-circulant matrices. In: AIMS Math. 8.7 (2023), pp. 17307–17334.], and by H. Huang, C. Li, and L. Deng [H. Huang, C. Li, and L. Deng. Public-Key Cryptography Based on Tropical Circular Matrices. In: Appl. Sci. 12.15 (2022), p. 7401] are insecure.

The impossible differential (ID) attack is crucial for analyzing the strength of block ciphers. The critical aspect of this technique is to identify IDs, and the researchers introduced several methods to detect them. Recently, the researchers extended the mixed-integer linear programming (MILP) approach by partitioning the input and output differences to identify IDs. The researchers proposed techniques to determine the representative set and partition table of a set over any nonlinear function. In this paper, we introduce a deterministic algorithm using the greedy approach~\cite{cormen2022introduction} for finding the representative set and partition table of a set over any nonlinear function. This algorithm iteratively selects the set that covers the maximum uncovered elements of the target set. This performs better than the existing algorithms in terms of time complexity. We use this algorithm to compute the representative sets and partition tables of the two block ciphers, IVLBC and GIFT-64, and identify 6-round IDs for them. Our research contributes a deterministic algorithm to compute the representative set and partition table of a set over any nonlinear function with at most 16-bit input size.

Recent work proposed by Bernstein et al. (from EPRINT 2024) identified two timing attacks, KyberSlash1 and KyberSlash2, targeting ML-KEM decryption and encryption algorithms, respectively, enabling efficient recovery of secret keys. To mitigate these vulnerabilities, correctives were promptly applied across implementations. In this paper, we demonstrate a very simple side-channel-assisted power analysis attack on the patched implementations of ML-KEM. Our result showed that original timing leakage can be shifted to power consumption leakage that can be exploited on specific data. We performed a practical validation of this attack on both the standard and a shuffled implementations of ML-KEM on a Cortex-M4 platform, confirming its effectiveness. Our approach enables the recovery of the ML-KEM secret key in just 30 seconds for the standard implementation, and approximately 3 hours for the shuffled implementation, achieving a 100% success rate in both cases.

The CRYSTALS-Dilithium digital signature scheme, selected by NIST as a post-quantum cryptography (PQC) standard under the name ML-DSA, employs a public key compression technique intended for performance optimization. Specifically, the module learning with error instance $({\bf A}, {\bf t})$ is compressed by omitting the low-order bits ${\bf t_0}$ of the vector ${\bf t}$. It was recently shown that knowledge of ${\bf t_0}$ enables more effective side-channel attacks on Dilithium implementations. Another recent work demonstrated a method for reconstructing ${\bf t_0}$ from multiple signatures. In this paper, we build on this method by applying profiled deep learning-assisted side-channel analysis to partially recover the least significant bit of ${\bf t_0}$ from power traces. As a result, the number of signatures required for the reconstruction of ${\bf t_0}$ can be reduced by roughly half. We demonstrate how the new ${\bf t_0}$ reconstruction method enhances the efficiency of recovering the secret key component ${\bf s}_1$, and thus facilitates digital signature forgery, on an ARM Cortex-M4 implementation of Dilithium.

In PKC'24, de Saint Guilhem and Pedersen give a pseudorandom function basing on a relaxed group action assumption in the semi-honest setting. Basing on the assumption, they build an oblivious pseudorandom function (OPRF). Later, a recent paper by Levin and Pedersen uses the same function to build a verifiable random function (VRF), using the same assumption. We give a structural attack on this problem by reducing it to a few group action inverse problems (GAIP/DLog) over small subgroups. This reduction allows us to apply a CRT-based attack to recover the secret key, ultimately lowering the problem’s effective security strength to under 70 classical bits when using CSIDH-512. Hence the strength of their pseudorandom functions is bounded above by the GAIP over the largest prime order subgroup. Clearly, Kuperberg’s subexponential attack can be used to further reduce its quantum security.

In this paper, we investigate the security of lightweight block ciphers, focusing on those that utilize the ADD-Rotate-XOR (ARX) and AND-Rotate-XOR (AND-RX) design paradigms. More specifically, we examine their resilience against boomerang-style attacks. First, we propose an automated search strategy that leverages the boomerang connectivity table (BCT) for AND operations ($\wedge BCT$) to conduct a complete search for boomerang and rectangle distinguishers for AND-RX ciphers. The proposed search strategy automatically considers all possible $\wedge BCT$ switches in the middle of the boomerang to optimise distinguishing probability. The correctness of the search strategy was verified experimentally. We were able to find the best boomerang and rectangle distinguishers to date in the single-key model for lightweight block ciphers KATAN32/48/64} and SIMON32/48. Next, we investigated BCT properties of ARX ciphers and discovered that a truncated boomerang switch could be formulated for the lightweight ARX cipher, CHAM. We were able to find the best single-key and related-key rectangle distinguishers to date for Cham. Our findings provide more accurate security margins of these lightweight ciphers against boomerang-style attacks.

This article presents a cryptanalysis of a 19th-century encrypted manuscript discovered in the archives of Conde de Siete Fuentes in Tenerife, Canary Islands, Spain. The manuscript, preserved by the heirs of the 6th Count of Valle de Salazar, utilizes a polyalphabetic substitution cipher. The cryptanalysis was performed by applying statistical frequency analysis and developing a Python script for decryption, resulting in the authors successfully deciphering the message. The decrypted letter reveals political communications discussing the strategic positioning of Tenerife as the capital, the dissolution of local councils, and the influence of key political figures. The analysis compares the cipher with historical encryption techniques, and identifies the unique characteristics of the manuscript’s encryption method. The study highlights the political dynamics and alliances within Tenerife’s nobility and their interactions with the central Spanish government, providing significant insights into, both, the cryptographic practices and political maneuvers of the time.

This work investigates persistent fault analysis on ASCON cipher that has been recently standardized by NIST USA for lightweight cryptography applications. In persistent fault, the fault once injected through RowHammer injection techniques, exists in the system during the entire encryption phase. In this work, we propose a model to mount persistent fault analysis (PFA) on ASCON cipher. In the finalization round of the ASCON cipher, we identify that the fault-injected S-Box operation in the permutation round, $p^{12}$, is vulnerable to leaking infor- mation about the secret key. The model can exist in two variants, a single instance of fault-injected S-Box out of 64 parallel S-Box invocations, and the same faulty S-Box iterated 64 times. The attack model demonstrates that any Spongent construction operating with authenticated encryption with associated data (AEAD) mode is vulnerable to persistent faults. In this work, we demonstrate the scenario of a single fault wherein the fault, once injected is persistent until the device is powered off. Using the pro- posed method, we successfully retrieve the 128-bit key in ASCON. Our experiments show that the minimum number and the maximum num- ber of queries required are 63 plaintexts and 451 plaintexts, respectively. Moreover, we observe that the number of queries required to mount the attack depends on fault location in the S-box LUT as observed from the plots reporting the minimum number of queries and average number of queries for 100 key values.

A Bitcoin miner who owns a sufficient amount of mining power can perform selfish mining to increase its relative revenue. Studies have demonstrated that the time-averaged profit of a selfish miner starts to rise once the mining difficulty level gets adjusted in favor of the attacker. Selfish mining profitability lies in the fact that orphan blocks are not incorporated into the current version of Bitcoin's difficulty adjustment mechanism (DAM). Therefore, it is believed that considering the count of orphan blocks in the DAM can result in complete unprofitability for selfish mining. In this paper, we disprove this belief by providing a formal analysis of the selfish mining time-averaged profit. We present a precise definition of the orphan blocks that can be incorporated into calculating the next epoch's target and then introduce two modified versions of DAM in which both main-chain blocks and orphan blocks are incorporated. We propose two versions of smart intermittent selfish mining, where the first one dominates the normal intermittent selfish mining, and the second one results in selfish mining profitability under the modified DAMs. Moreover, we present the orphan exclusion attack with the help of which the attacker can stop honest miners from reporting the orphan blocks. Using combinatorial tools, we analyze the profitability of selfish mining accompanied by the orphan exclusion attack under the modified DAMs. Our results show that even when considering orphan blocks in the DAM, selfish mining can still be profitable. However, the level of profitability under the modified DAMs is significantly lower than that observed under the current version of Bitcoin DAM, suggesting that orphan reporting can be an effective countermeasure against a payoff-maximizing selfish miner.

This study aims to determine the complete and precise differential properties of SM4, which have remained unknown for over twenty years after the cipher was initially released. A Boolean Satisfiability Problem (SAT) based automatic search approach is employed to achieve the objective. To improve the limited efficiency of the search focused on differential probabilities, we want to investigate the feasibility of integrating human expertise into an automatic approach to enhance the search speed. This study presents the construction of four new SAT models that describe the human-identified specific properties of short differential characteristics. All of these models are integrated into the fundamental model, and the SAT solver is implemented to assess the acceleration capabilities of the new models. The experimental results indicate that including three new models effectively decreases the overall execution time of the SAT solver. Using the novel models, we obtain the first precise minimal values for the number of active S-boxes of SM4 under single-key (complete rounds) and related-key (1-round to 19-round) settings. The first precise upper bound for differential probabilities of SM4 (1-round to 20-round) is also determined. In addition, we present the first publicly revealed optimal 19-round differential characteristic of SM4.

In CRYPTO 2019, Gohr introduced the method of differential neural cryptanalysis, utilizing neural networks as the underlying distinguishers to achieve distinguishers for (5-8)-round of the Speck32/64 cipher and subsequently recovering keys for 11 and 12 rounds. Inspired by this work, we propose an enhanced neural cryptanalysis framework that combines the Efficient Channel Attention (ECA) module with residual networks. By introducing the channel attention mechanism to emphasize key features and leveraging residual networks to facilitate efficient feature extraction and gradient flow, we achieve improved performance. Additionally, we employ a new data format that combines the ciphertext and the penultimate round ciphertext as input samples, providing the distinguisher with more useful features. Compared with the known results, our work enhance the accuracy of the neural distinguishers for Simeck32/64 (10-12)-round and achieve a new 13-round distinguisher. We also improve the accuracy of the Simeck48/96 (10-11)-round distinguishers and develop new (12-16)-round neural distinguishers. Moreover, we enhance the accuracy of the Simeck64/128 (14-18)-round distinguishers and obtain a new 19-round neural distinguisher. As a result, we achieve the highest accuracy and the longest rounds distinguishers for Simeck32/64, Simeck48/96, and Simeck64/128.

Oblivious pseudorandom function (OPRF) is a two-party cryptographic protocol that allows the receiver to input $x$ and learn $F(x)$ for some PRF $F$, only known to the sender. For private set intersection (PSI) applications, OPRF protocols have evolved to enhance efficiency, primarily using symmetric key cryptography. Current state-of-the-art protocols, such as those by Rindal and Schoppmann (Eurocrypt '21), leverage vector oblivious linear evaluation (VOLE) and oblivious key-value store (OKVS) constructions. In this work, we identify a flaw in an existing security proof, and present practical attacks in the malicious model, which results in additional PRF evaluations than the previous works' claim. In particular, the attack for malicious model is related to the concept of OKVS overfitting, whose hardness is conjectured in previous works. Our attack is the first one to discuss the concrete hardness of OKVS overfitting problem. As another flavour of contribution, we generalize OKVS-based OPRF constructions, suggesting new instantiations using a VOLE protocol with only Minicrypt assumptions. Our generalized construction shows improved performance in high-speed network environments, narrowing the efficiency gap between the OPRF constructions over Cryptomania and Minicrypt.

The rapid development of advanced cryptographic applications like multi-party computation (MPC), fully homomorphic encryption (FHE), and zero-knowledge (ZK) proofs have motivated the designs of the so-called arithmetic-oriented (AO) primitives. Efficient AO primitives typically build over large fields and use large S-boxes. Such design philosophy brings difficulties in the cryptanalysis of these primitives as classical cryptanalysis methods do not apply well. The generally recognized attacks against these primitives are algebraic attacks, especially Groebner basis attacks. Thus, the numbers of security rounds are usually derived through the complexity of solving the system of algebraic equations using Groebner bases. In this paper, we propose a novel framework for algebraic attacks against AO primitives. Instead of using Groebner basis, we use resultants to solve a system of multivariate equations that can better exploit the algebraic structures of AO primitives. We employ several techniques to reduce the dimensions of the resultants and avoid rapid increases in degrees, including meet-in-the-middle modeling, variable substitutions, and fast Lagrange interpolation. We apply our attack to three mainstream AO cryptographic primitives: Rescue-Prime, Anemoi, and Jarvis. For Rescue-Prime, we theoretically prove that the final univariate equation has a degree of at most a specific power of three and practically attack five rounds for the first time. We attack the full-round of Anemoi with complexity 2^110.10, which has been claimed to provide 127 bits of security. We also give the first practical attack against eight rounds of Anemoi over a 55-bit prime field. For Jarvis, we improve the existing practical attack by a factor of 100. Therefore, we point out that our analysis framework can be used as a new evaluation method for AO designs.

Voltage fault injection attacks are a particularly powerful threat to secure embedded devices because they exploit brief, hard-to-detect power fluctuations causing errors or bypassing security mechanisms. To counter these attacks, various detectors are employed, but as defenses strengthen, increasingly elusive glitches continue to emerge. Artificial intelligence, with its inherent ability to learn and adapt to complex patterns, presents a promising solution. This research presents an AI-driven voltage fault injection detector that analyzes clock signals directly. We provide a detailed fault characterization of the STM32F410 microcontroller, emphasizing the impact of faults on the clock signal. Our findings reveal how power supply glitches directly impact the clock, correlating closely with the amount of power injected. This led to developing a lightweight Multi-Layer Perceptron model that analyzes clock traces to distinguish between safe executions, glitches that keep the device running but may introduce faults, and glitches that cause the target to reset. While previous fault injection AI applications have primarily focused on parameter optimization and simulation assistance, in this work we use the adaptability of machine learning to create a fault detection model that is specifically adjusted to the hardware that implements it. The developed glitch detector has a high accuracy showing this a promising direction to combat FI attacks on a variety of platform.

We develop an efficient algorithm to detect whether a superspecial genus 2 Jacobian is optimally $(N, N)$-split for each integer $N \leq 11$. Incorporating this algorithm into the best-known attack against the superspecial isogeny problem in dimension 2 gives rise to significant cryptanalytic improvements. Our implementation shows that when the underlying prime $p$ is 100 bits, the attack is sped up by a factor $25{\tt x}$; when the underlying prime is 200 bits, the attack is sped up by a factor $42{\tt x}$; and, when the underlying prime is 1000 bits, the attack is sped up by a factor $160{\tt x}$.

Computationally hard problems based on coding theory, such as the syndrome decoding problem, have been used for constructing secure cryptographic schemes for a long time. Schemes based on these problems are also assumed to be secure against quantum computers. However, these schemes are often considered impractical for real-world deployment due to large key sizes and inefficient computation time. In the recent call for standardization of additional post-quantum digital signatures by the National Institute of Standards and Technology, several code-based candidates have been proposed, including LESS, CROSS, and MEDS. These schemes are designed on the relatively new zero-knowledge framework. Although several works analyze the hardness of these schemes, there is hardly any work that examines the security of these schemes in the presence of physical attacks. In this work, we analyze these signature schemes from the perspective of fault attacks. All these schemes use a similar tree-based construction to compress the signature size. We attack this component of these schemes. Therefore, our attack is applicable to all of these schemes. In this work, we first analyze the LESS signature scheme and devise our attack. Furthermore, we showed how this attack can be extended to the CROSS signature scheme. Our attacks are built on very simple fault assumptions. Our results show that we can recover the entire secret key of LESS and CROSS using as little as a single fault. Finally, we propose various countermeasures to prevent these kinds of attacks and discuss their efficiency and shortcomings.

BAKSHEESH is a lightweight block cipher following up the well-known cipher GIFT-128, which uses a 4-bit SBox that has a non-trivial Linear Structure (LS). Also, the Sbox requires a low number of AND gates that makes BAKSHEESH stronger to resist the side channel attacks compared to GIFT-128. In this paper, we give the first third-party security analysis of BAKSHEESH from the traditional attacks perspective: integral, differential and linear attacks. Firstly, we propose a framework for integral attacks based on the properties of BAKSHEESH's Sbox and its inverse. By this, we achieve the 9- and 10-round practical key-recovery attacks, and give a 15-round theoretical attack. Secondly, we re-evaluate the security bound against differential cryptanalysis, correcting two errors from the original paper and presenting a key-recovery attack for 19 rounds. At last, for linear cryptanalysis, we develop an automated model for key-recovery attacks and then demonstrate a key-recovery attack for 21 rounds. We stress that our attacks cannot threaten the full-round BAKSHEESH, but give a deep understanding on its security.

The Linear Code Equivalence (LCE) Problem has received increased attention in recent years due to its applicability in constructing efficient digital signatures. Notably, the LESS signature scheme based on LCE is under consideration for the NIST post-quantum standardization process, along with the MEDS signature scheme that relies on an extension of LCE to the rank metric, namely the Matrix Code Equivalence (MCE) Problem. Building upon these developments, a family of signatures with additional properties, including linkable ring, group, and threshold signatures, has been proposed. These novel constructions introduce relaxed versions of LCE (and MCE), wherein multiple samples share the same secret equivalence. Despite their significance, these variations have often lacked a thorough security analysis, being assumed to be as challenging as their original counterparts. Addressing this gap, our work delves into the sample complexity of LCE and MCE --- precisely, the sufficient number of samples required for efficient recovery of the shared secret equivalence. Our findings reveal, for instance, that one should not use the same secret twice in the LCE setting since this enables a polynomial time (and memory) algorithm to retrieve the secret. Consequently, our results unveil the insecurity of two advanced signatures based on variants of the LCE Problem.

In 2022, Wang et al. proposed the multivariate signature scheme SNOVA as a UOV variant over the non-commutative ring of $\ell \times \ell $ matrices over $\mathbb{F}_q$. This scheme has small public key and signature size and is a first round candidate of NIST PQC additional digital signature project. Recently, Ikematsu and Akiyama, and Li and Ding show that the core matrices of SNOVA with $v$ vinegar-variables and $o$ oil-variables are regarded as the representation matrices of UOV with $\ell v$ vinegar-variables and $\ell o$ oil-variables over $\mathbb{F}_q$, and thus we can apply existing key recovery attacks as a plain UOV. In this paper, we propose a method that reduces SNOVA to smaller UOV with $v$ vinegar-variables and $o$ oil-variables over $\mathbb{F}_{q^\ell }$. As a result, we show that the previous first round parameter sets at $\ell = 2$ do not meet the NIST PQC security levels. We also confirm that the present parameter sets are secure from existing key recovery attacks with our approach.

A new design strategy for ZK-friendly hash functions has emerged since the proposal of $\mathsf{Reinforced Concrete}$ at CCS 2022, which is based on the hybrid use of two types of nonlinear transforms: the composition of some small-scale lookup tables (e.g., 7-bit or 8-bit permutations) and simple power maps over $\mathbb{F}_p$. Following such a design strategy, some new ZK-friendly hash functions have been recently proposed, e.g., $\mathsf{Tip5}$, $\mathsf{Tip4}$, $\mathsf{Tip4}'$ and the $\mathsf{Monolith}$ family. All these hash functions have a small number of rounds, i.e., $5$ rounds for $\mathsf{Tip5}$, $\mathsf{Tip4}$, and $\mathsf{Tip4}'$, and $6$ rounds for $\mathsf{Monolith}$ (recently published at ToSC 2024/3). Using the composition of some small-scale lookup tables to build a large-scale permutation over $\mathbb{F}_p$ - which we call S-box - is a main feature in such designs, which can somehow enhance the resistance against the Gröbner basis attack because this large-scale permutation will correspond to a complex and high-degree polynomial representation over $\mathbb{F}_p$. As the first technical contribution, we propose a novel and efficient algorithm to study the differential property of this S-box and to find a conforming input pair for a randomly given input and output difference. For comparison, a trivial method based on the use of the differential distribution table (DDT) for solving this problem will require time complexity $\mathcal{O}(p^2)$. For the second contribution, we also propose new frameworks to devise efficient collision attacks on such hash functions. Based on the differential properties of these S-boxes and the new attack frameworks, we propose the first collision attacks on $3$-round $\mathsf{Tip5}$, $\mathsf{Tip4}$, and $\mathsf{Tip4}'$, as well as $2$-round $\mathsf{Monolith}$-$31$ and $\mathsf{Monolith}$-$64$, where the $2$-round attacks on $\mathsf{Monolith}$ are practical. In the semi-free-start (SFS) collision attack setting, we achieve practical SFS collision attacks on $3$-round $\mathsf{Tip5}$, $\mathsf{Tip4}$, and $\mathsf{Tip4}'$. Moreover, the SFS collision attacks can reach up to $4$-round $\mathsf{Tip4}$ and $3$-round $\mathsf{Monolith}$-$64$. As far as we know, this is the first third-party cryptanalysis of these hash functions, which improves the initial analysis given by the designers.

The Lattice Isomorphism Problem (LIP) was recently introduced as a new hardness assumption for post-quantum cryptography. The strongest known efficiently computable invariant for LIP is the genus of a lattice. To instantiate LIP-based schemes one often requires the existence of a lattice that (1) lies in some fixed genus, and (2) has some good geometric properties such as a high packing density or small smoothness parameter. In this work we show that such lattices exist. In particular, building upon classical results by Siegel (1935), we show that essentially any genus contains a lattice with a close to optimal packing density, smoothing parameter and covering radius. We present both how to efficiently compute concrete existence bounds for any genus, and asymptotically tight bounds under weak conditions on the genus.

Interconnected devices enhance daily life but introduce security vulnerabilities, new technologies enable malicious activities such as information theft. This article combines radio frequency (RF) side-channel attacks with software Trojans to create a hard-to-detect, stealthy method for extracting kilobytes of secret information per millisecond over record distances with a single measurement in the RF spectrum. The technique exploits Trojan-induced electrical disturbances in RF components originating from peripherals, buses, memories and CPUs to achieve high SNR data leakage schemes. Experimental results show negligible acquisition time and stealth. The research introduces optimized modulation, demodulation schemes, and specialized synchronization symbols to minimize error rates and maximize data rates. It highlights the need for advanced detection and defense mechanisms to ensure the security and privacy of interconnected devices.

Impossible differential cryptanalysis is a crucial cryptanalytical method for symmetric ciphers. Given an impossible differential, the key recovery attack typically proceeds in two steps: generating pairs of data and then identifying wrong keys using the guess-and-filtering method. At CRYPTO 2023, Boura \etal first proposed a new key recovery technique - the differential meet-in-the-middle attack, which recovers the key in a meet-in-the-middle manner. Inspired by this technique, we incorporate the meet-in-the-middle technique into impossible cryptanalysis and propose a generic impossible differential meet-in-the-middle attack (\idma) framework. We apply \idma to block ciphers \skinny, \skinnye-v2, and \forkskinny and achieve remarkably efficient attacks. We improve the impossible differential attack on \skinny-$n$-$3n$ by 2 rounds in the single-tweakey setting and 1 round in the related-tweakey setting. For \skinnye-v2, the impossible differential attacks now can cover 2 more rounds in the related-tweakey setting and the first 23/24/25-round attacks in the single-tweakey model are given. For \forkskinny-$n$-$3n$, we improve the attacks by 2 rounds in the limited setting specified by the designers and 1 round in relaxed settings. These results confirm that the meet-in-the-middle technique can result in more efficient key recovery, reaching beyond what traditional methods can achieve on certain ciphers.

Gröbner basis cryptanalysis of hash functions and ciphers, and their underlying permutations, has seen renewed interest recently. Anemoi (Crypto'23) is a permutation-based hash function that is efficient for a variety of arithmetizations used in zero-knowledge proofs. In this paper, exploring both theoretical bounds as well as experimental validation, we present new complexity estimates for Gröbner basis attacks on the Anemoi permutation over prime fields. We cast our findings in what we call the six worlds of Gröbner basis cryptanalysis. As an example, keeping the same security arguments of the design, we conclude that at least 41 instead of 37 rounds would need to be used for 256-bit security, whereby our suggestion does not yet include a security margin.

Implementation-security vulnerabilities such as the power-based side-channel leakage and fault-injection sensitivity of a secure chip are hard to verify because of the sophistication of the measurement setup, as well as the need to generalize the adversary into a test procedure. While the literature has proposed a wide range of vulnerability metrics to test the correctness of a secure implementation, it is still up to the subject-matter expert to map these concepts into a working and reliable test procedure. Recently, we investigated the benefits of using an open-source implementation security testing environment called Chipwhisperer. The open-source and low-cost nature of the Chipwhisperer hardware and software has resulted in the adoption of thousands of testing kits throughout academia and industry, turning the testkit into a baseline for implementation security testing. We investigate the use cases for the Chipwhisperer hardware and software, and we evaluate the feasibility of an open-source ecosystem for implementation security testing. In addition to the open-source hardware and firmware, an ecosystem also considers broader community benefits such as re-usability, sustainability, and governance.

Private set intersections are cryptographic protocols that compute the intersection of multiple parties' private sets without revealing elements that are not in the intersection. These protocols become less efficient when the number of parties grows, or the size of the sets increases. For this reason, many protocols are based on Bloom filters, which speed up the protocol by approximating the intersections, introducing false positives with a small but non-negligible probability. These false positives are caused by hash collisions in the hash functions that parties use to encode their sets as Bloom filters. In this work, we show that these false positives are more than an inaccuracy: an adversary in the augmented semi-honest model can use them to learn information about elements that are not in the intersection. First, we show that existing security proofs for Bloom filter-based private set intersections are flawed. Second, we show that even in the most optimistic setting, Bloom filter-based private set intersections cannot securely realize an approximate private set intersection unless the parameters are so large that false positives only occur with negligible probability. Third, we propose a practical attack that allows a party to learn if an element is contained in a victim's private set, showing that the problem with Bloom filters is not just theoretical. We conclude that the efficiency gain of using Bloom filters as an approximation in existing protocols vanishes when accounting for this security problem. We propose three mitigations besides choosing larger parameters: One can use oblivious pseudo-random functions instead of hash functions to reduce the success rate of our attack significantly, or replace them with password-based key derivation functions to significantly slow down attackers. A third option is to let a third party authorize the input sets before proceeding with the protocol.

In the face of escalating security threats in modern computing systems, there is an urgent need for comprehensive defense mechanisms that can effectively mitigate invasive, noninvasive and interactive security vulnerabilities in hardware and software domains. Individually, hardware and software weaknesses and probable remedies have been practiced but protecting a combined system has not yet been discussed in detail. This survey paper provides a comprehensive overview of the emerging field of Hardware-Software co-Protection against Invasive and Non-Invasive Security Threats. We systematically review state-of-the-art research and developments in hardware and software security techniques, focusing on their integration to create synergistic defense mechanisms. The survey covers a wide range of security threats, including physical attacks, side-channel attacks, and malware exploits, and explores the diverse strategies employed to counter them. Our survey meticulously examines the landscape of security vulnerabilities, encompassing both physical and software-based attack vectors, and explores the intricate interplay between hardware and software defenses in mitigating these threats. Furthermore, we discuss the challenges and opportunities associated with Hardware-Software co-Protection and identify future research directions to advance the field. Through this survey, we aim to provide researchers, practitioners, and policymakers with valuable insights into the latest advancements and best practices for defending against complex security threats in modern computing environments.

Chosen-prefix collision (CPC) attack was first presented by Stevens, Lenstra and de Weger on MD5 at Eurocrypt 2007. A CPC attack finds a collision for any two chosen prefixes, which is a stronger variant of collision attack. CPCs are naturally harder to construct but have larger practical impact than (identical-prefix) collisions, as seen from the series of previous works on MD5 by Stevens et al. and SHA-1 by Leurent and Peyrin. Despite its significance, the resistance of CPC attacks has not been studied on AES-like hashing. In this work, we explore CPC attacks on AES-like hashing following the framework practiced on MD5 and SHA-1. Instead of the message modification technique developed for MD-SHA family, we opt for related-key rebound attack to construct collisions for AES-like hashing in view of its effectiveness. We also note that the CPC attack framework can be exploited to convert a specific class of one-block free-start collisions into two-block collisions, which sheds light on the importance of free-start collisions. As a result, we present the first CPC attacks on reduced Whirlpool, Saturnin-hash and AES-MMO/MP in classic and quantum settings, and extend the collision attack on Saturnin-hash from 5 to 6 rounds in the classic setting. As an independent contribution, we improve the memoryless algorithm of solving 3-round inbound phase by Hosoyamada and Sasaki at Eurocrpyt 2020, which leads to improved quantum attacks on Whirlpool. Notably, we find the first 6-round memoryless quantum collision attack on Whirlpool better than generic CNS collision finding algorithm when exponential-size qRAM is not available but exponential-size classic memory is available.

SLIM and LBCIoT are lightweight block ciphers proposed for IoT applications. We present differential meet-in-the-middle attacks on these ciphers and discuss several implementation variants and possible improvements of these attacks. Experimental validation also shows some results that may be of independent interest in the cryptanalysis of other ciphers. Namely, the problems with low-probability differentials and the questionable accuracy of standard complexity estimates of using filters.

In this paper, we present the first single trace side-channel attack that targets the MPC-in-the-Head (MPCitH) framework based on threshold secret sharing, also known as Threshold Computation in the Head (TCitH) in its original version. This MPCitH framework can be found in 5 of the 14 digital signatures schemes in the recent second round of the National Institute of Standards and Technology (NIST) call for digital signatures. In this work, we start by highlighting a side-channel vulnerability of the TCitH framework and show an exploitation of it on the SDitH algorithm, which is part of this NIST call. Specifically, we exploit the leakage of a multiplication function in the Galois field to make predictions about intermediate values, and we use the structure of the algorithm to combine information efficiently. This allows us to build an attack that is both the first Soft Analytical Side-Channel Attack (SASCA) targeting the MPCitH framework, as well as the first attack on SDitH. More specifically, we build a SASCA based on Belief Propagation (BP) on the evaluation of polynomials in the signature using the threshold variant structure to reconstruct the secret key. We perform simulated attacks under the Hamming Weight (HW) leakage model, enabling us to evaluate the resistance of the scheme against SASCA. We then perform our attacks in a real case scenario, more specifically on the STM32F407, and recover the secret key for all the security levels. We end this paper by discussing the various shuffling countermeasures we could use to mitigate our attacks.

The implementation security of post-quantum cryptography (PQC) algorithms has emerged as a critical concern with the PQC standardization process reaching its end. In a side-channel-assisted chosen-ciphertext attack, the attacker builds linear inequalities on secret key components and uses the belief propagation (BP) algorithm to solve. The number of inequalities leverages the query complexity of the attack, so the fewer the better. In this paper, we use the PQC standard algorithm CRYSTALS-Kyber as a study case to construct bilateral inequalities on key variables with substantially narrower intervals using a side-channel-assisted oracle. For Kyber512, Kyber768, and Kyber 1024, the average Shannon entropy carried by such inequality is improved from the previous 0.6094, 0.4734, and 0.8544 to 0.6418, 0.4777, and 1.2007. The number of such inequalities required to recover the key utilizing the BP algorithm for Kyber512 and Kyber1024 is reduced by 5.32% and 40.53% in theory and experimentally the reduction is even better. The query complexity is reduced by 43%, 37%, and 48% for Kyber512, 768, and 1024 assuming reasonably perfect reliability. Furthermore, we introduce a strategy aimed at further refining the interval of inequalities. Diving into the BP algorithm, we discover a measure metric named JSD (Jensen-Shannon distance)-metric that can gauge the tightness of an inequality. We then develop a machine learning-based strategy to utilize the JSD-metrics to contract boundaries of inequalities even with fewer inequalities given, thus improving the entropy carried by the system of linear inequalities. This contraction strategy is at the algorithmic level and has the potential to be employed in all attacks endeavoring to establish a system of inequalities concerning key variables.

The impossible boomerang (IB) attack was first introduced by Lu in his doctoral thesis and subsequently published at DCC in 2011. The IB attack is a variant of the impossible differential (ID) attack by incorporating the idea of the boomerang attack. In this paper, we revisit the IB attack, and introduce the incompatibility of two characteristics in boomerang to the construction of an IB distinguisher. With our methodology, all the constructions of IB distinguisher are represented in a unified manner. Moreover, we show that the related-(twea)key IB distinguishers possess more freedom than the ones of ID so that it can cover more rounds. We also propose a new tool based on Mixed-Integer Quadratically-Constrained Programming (MIQCP) to search for IB attacks. To illustrate the power of the IB attack, we mount attacks against three tweakable block ciphers: Deoxys-BC, Joltik-BC and SKINNY. For Deoxys-BC, we propose a related-tweakey IB attack on 14-round Deoxys-BC-384, which improves the best previous related-tweakey ID attack by 2 rounds, and we improve the data complexity of the best previous related-tweakey ID attack on 10-round Deoxys-BC-256. For Joltik-BC, we propose the best attacks against 10-round Joltik-BC-128 and 14-round Joltik-BC-192 with related-tweakey IB attack. For SKINNY-n-3n, we propose a 27-round related-tweakey IB attack, which improves both the time and the memory complexities of the best previous ID attack. We also propose the first related-tweakey IB attack on 28 round SKINNY-n-3n, which improves the previous best ID attack by one round.

Pseudo-random generators are deterministic algorithms that take in input a random secret seed and output a flow of random-looking numbers. The Knapsack generator, presented by Rueppel and Massey in 1985 is one of the many attempt at designing a pseudo-random generator that is cryptographically secure. It is based on the subset-sum problem, a variant of the Knapsack optimization problem, which is considered computationally hard. In 2011 Simon Knellwolf et Willi Meier found a way to go around this hard problem and exhibited a weakness of this generator. In addition to be able to distinguish the outputs from the uniform distribution, they designed an algorithm that retrieves a large portion of the secret. We present here an alternate version of the attack, with similar costs, that works on a larger range of parameters and retrieves a larger portion of the secret.

Boneh et al. (CRYPTO'18) proposed two $t$-out-of-$N$ threshold fully homomorphic encryption ($\sf TFHE$) schemes based on Shamir secret sharing scheme and $\{0,1\}$-linear secret sharing scheme. They demonstrated the simulation security, ensuring no information leakage during partial or final decryption. This breakthrough allows any scheme to be converted into a threshold scheme by using $\sf TFHE$. We propose two polynomial time algorithms to break the simulation security of $t$-out-of-$N$ $\sf TFHE$ based on Shamir secret sharing scheme proposed by Boneh et al.. First, we show that an adversary can break the simulation security by recovering the secret key under some constraints on $t$ and $N$, which does not violate the conditions for security proof. Next, we introduce a straightforward fix that theoretically satisfies the simulation security. However, we argue that this modification remains insecure insecure when implemented with any state-of-the-art fully homomorphic encryption libraries in practice. To ensure robustness against our subsequent attacks, we recommend using an error-refreshing algorithm, such as bootstrapping or modulus switching, for each addition operation.

NTRU-like constructions are among the most studied lattice-based schemes. The freedom of design of NTRU resulted in many variants in literature motivated by faster computations or more resistance against lattice attacks by changing the underlying algebra. To the best of our knowledge, BQTRU (DCC 2017), a noncommutative NTRU-like cryptosystem, is the fastest claimed variant of NTRU built over the quaternion algebra of the bivariate ring of polynomials. The key generation and the encryption of BQTRU are claimed to be 16/7 times faster than standard NTRU for equivalent levels of security. For key recovery attacks, the authors claim that retrieving a decryption key is equivalent to solving the Shortest Vector Problem (SVP) in expanded Euclidean lattices of giant dimensions. This work disproves this claim and proposes practical key and message recovery attacks that break the moderate parameter sets of BQTRU estimated to achieve $2^{92}$ message security and $2^{166}$ key security on a standard desktop within less than two core weeks. Furthermore, our analysis shows that the proposed parameter set for the highest security level claiming $2^{212}$ message security and $2^{396}$ key security can barely achieve $2^{82}$ message security and $2^{125}$ key security. Our work not only provides cryptanalysis for BQTRU but also demonstrates the potential of extending Gentry's attack to other rings beyond the cyclotomic polynomial ring.

Last year CRYSTALS-Kyber was chosen by NIST as a new, post-quantum secure key encapsulation mechanism to be standardized. This makes it important to assess the resistance of CRYSTALS-Kyber implementations to physical attacks. Pure side-channel attacks on post-quantum cryptographic algorithms have already been well-explored. In this paper, we present an attack on a masked and shuffled software implementation of CRYSTALS-Kyber that combines fault injection with side-channel analysis. First, a voltage fault injection is performed to bypass the shuffling. We found settings that consistently glitch the desired instructions without crashing the device. After the successful fault injection, a deep learning-assisted profiled power analysis based on the Hamming weight leakage model is used to recover the message (shared key). We propose a partial key enumeration method that allows us to significantly increase the success rate of message recovery (from 0.122 without enumeration to 0.887 with 32 enumerated bits).

CRYSTALS-Dilithium is a post-quantum secure digital signature algorithm currently being standardised by NIST. As a result, devices making use of CRYSTALS-Dilithium will soon become generally available and be deployed in various environments. It is thus important to assess the resistance of CRYSTALS-Dilithum implementations to physical attacks. In this paper, we present an attack on a CRYSTALS-Dilithium implementation in hedged mode in ARM Cortex-M4 using fault injection. Voltage glitching is performed to skip computation of a seed during the generation of the signature. We identified settings that consistently skip the desired function without crashing the device. After the successful fault injection, the resulting signature allows for the extraction of the secret key vector. Our attack succeeds with probability 0.582 in a single trace. We also propose countermeasures against the presented attack.

The analytical perspective employed in the study classifies the theoretical research on dependencies in differential characteristics into two types. By categorising all dependence representations from the value restrictions and the theory of quasidifferential trails, we pinpoint a specific set of nonlinear constraints, which we term linearised nonlinear constraints. We aim to establish a method that utilises value restrictions to identify these constraints, as the current method based on value restrictions is found to be lacking in this area. A linearisation method for searching linearised nonlinear constraints for a given differential characteristic is developed by leveraging linear dependencies between inputs and outputs of active S-boxes. Then, we propose a three-stage evaluation approach to more accurately evaluate differential characteristics with linearised nonlinear constraints. Four differential characteristics of GIFT-64 are analysed using the three-stage evaluation approach, and the exact right key spaces and remaining probabilities are given. According to our results, the right key spaces of the four differential characteristics do not cover the entire key space, and the remaining probabilities are not equivalent to the stated probabilities. Concerning GIFT-128, we find six differential characteristics subject to linearised nonlinear constraints. Besides, inconsistencies are detected in the linear and linearised nonlinear constraints in the characteristics of two differentials employed to initiate the most effective differential attack on GIFT-128. Based on these results, we strongly advise reassessing the differential attacks that rely on these distinguishers. An additional advantage of using the linearisation method and the three-stage evaluation approach is their ability to identify linear and nonlinear constraints in ciphers that utilise the Generalised Feistel Network (GFN). It leads to the first instantiations of linear and nonlinear constraints in the GFN cipher WARP.

Remote Keyless Entry (RKE) systems are ubiqui- tous in modern day automobiles, providing convenience for vehicle owners - occasionally at the cost of security. Most automobile companies have proprietary implementations of RKE; these are sometimes built on insecure algorithms and authentication mechanisms. This paper presents a compre- hensive study conducted on the RKE systems of multiple cars from four automobile manufacturers not previously explored. Specifically, we analyze the design, implementation, and security levels of 7 different cars manufactured by Honda, Maruti-Suzuki, Toyota, and Mahindra. We also do a deep dive into the RKE system of a particular Honda model. We evaluate the susceptibility of these systems to known vulnerabilities (such as RollJam and RollBack at- tacks). This is accomplished using a novel tool – ‘Puck- py’, that helps analyze RKE protocols. Our tool automates several aspects of the protocol analysis process, reducing time and logistical constraints in RKE research; we provide standardized protocols to execute various attacks using our Puck-Py tool. We find that, despite having a long period of time to fix security issues, several popular automobiles remain susceptible to attacks, including the basic RollJam attack.

In this paper we study search problems that arise very often in cryptanalysis: nested search problems, where each search layer has known degrees of freedom and/or constraints. A generic quantum solution for such problems consists of nesting Grover's quantum search algorithm or amplitude amplification (QAA) by Brassard et al., obtaining up to a square-root speedup on classical algorithms. However, the analysis of nested Grover or QAA is complex and introduces technicalities that in previous works are handled in a case-by-case manner. Moreover, straightforward nesting of $\ell$ layers multiplies the complexity by a constant factor $(\pi/2)^\ell$. In this paper, we aim to remedy both these issues and introduce a generic framework and tools to transform a classical nested search into a quantum procedure. It improves the state-of-the-art in three ways: 1) our framework results in quantum procedures that are significantly simpler to describe and analyze; 2) it reduces the overhead factor from $(\pi/2)^\ell$ to $\sqrt{\ell}$; 3) it is simpler to apply and optimize, without needing manual quantum analysis. We give generic complexity formulas and show that for concrete instances, numerical optimizations enable further improvements, reducing even more the gap to an exact quadratic speedup. We demonstrate our framework by giving a tighter analysis of quantum attacks on reduced-round AES.

High-performance crypto-engines have become crucial components in modern System-On-Chip (SoC) architectures across platforms, from servers to edge-IoTs’. Alas, their secure operation faces a significant obstacle caused by information-leakage accessed through Side-Channel Analysis (SCA). Adversaries exploit statistical-analysis techniques on measured (e.g.,) power and timing signatures generated during (e.g.,) encryption, extracting secrets. Mathematical countermeasures against such attacks often impose substantial power-performance-area overheads. Dynamic Voltage and Frequency Scaling (DVFS) techniques provide power-efficiency by varying power consumption according to workload; these modulations are called power-states. Unintentionally, DVFS introduces new inherent weaknesses exploitable by malicious actors: power-states leak information in both power and timing side-channels, measurable in software and hardware. We introduce a method to increase side-channel resistance using integrated voltage regulators and DVFS: (1) Pushing known prior-art in the topic to Ultra Low Power (ULP) regime (2) For the first time introducing a mechanism to aid in counteracting the inherent weakness of DVFS in SCA (3) Providing measurements performed on 40nm process ULP PLS15 test-chip down at 580 mV power-supply (4) Offering improved and parameterized resistance to remote-timing vulnerabilities inherent to DVFS. We present various results and perform a detailed analysis while comparing performance and security to prior-art. Importantly, our solution is configurable in terms of security, maintaining degrees-of-freedom for power-optimization of DVFS.

Recently, Picnic3 has introduced several alternative LowMC instances, which prompts the cryptanalysis competition for LowMC. In this paper, we provide new solutions to the competition with full S-box layers under single-data complexity. First, we present a new guess-and-determine attack framework that achieves the best trade-off in complexity, while effectively enhancing two algorithms applicable to 2-round LowMC cryptanalysis. Next, we present a new meet-in-the-middle attack framework for 2-/3-round LowMC, which can gradually reduce the number of variables and narrow down the range of candidate keys in stages. As a result, our 3-stage MITM attacks have both lower time complexity and memory complexity than the best previous 2-round attacks proposed by Banik et al. at ASIACRYPT 2021, with memory reduced drastically by a factor of $ 2^{29.7} \sim 2^{70.4} $.

Public key cryptography can be based on integer factorization and the discrete logarithm problem (DLP), applicable in multiplicative groups and elliptic curves. Regev’s recent quantum algorithm was initially designed for the factorization and was later extended to the DLP in the multiplicative group. In this article, we further extend the algorithm to address the DLP for elliptic curves. Notably, based on celebrated conjectures in Number Theory, Regev’s algorithm is asymptotically faster than Shor’s algorithm for elliptic curves. Our analysis covers all cases where Regev’s algorithm can be applied. We examine the general framework of Regev’s algorithm and offer a geometric description of its parameters. This preliminary analysis enables us to certify the success of the algorithm on a particular instance before running it. In the case of integer factorization, we demonstrate that there exists an in- finite family of RSA moduli for which the algorithm always fails. On the other hand, when the parameters align with the Gaussian heuristics, we prove that Regev’s algorithm succeeds. By noting that the algorithm naturally adapts to the multidimensional DLP, we proved that it succeeds for a certain range of parameters.

The Feistel structure represents a fundamental architectural component within the domain of symmetric cryptographic algorithms, with a substantial body of research conducted within the context of classical computing environments. Nevertheless, research into specific symmetric cryptographic algorithms utilizing the Feistel structure is relatively scarce in quantum computing environments. This paper builds upon a novel 4-round distinguisher proposed by Ito et al. for the Feistel structure under the quantum chosen-ciphertext attack (qCCA) setting. It introduces a 5-round distinguisher for Camellia. The efficacy of the distinguisher has been empirically validated. Furthermore, this paper combines Grover's algorithm with Simon's algorithm, utilizing an analysis of Camellia's key scheduling characteristics to construct a 9-round key recovery attack on Camellia algorithm. The time complexity for acquiring the correct key bits is $2^{61.5}$, and it requires 531 quantum bits. This represents the inaugural chosen-ciphertext attack on Camellia under the Q2 model.

This paper aims to provide a more comprehensive understanding of the optimal linear characteristics of BAKSHEESH. Initially, an explicit formula for the absolute correlation of the $R$-round optimal linear characteristic of BAKSHEESH is proposed when $R \geqslant 12$. By examining the linear characteristics of BAKSHEESH with three active S-boxes per round, we derive some properties of the three active S-boxes in each round. Furthermore, we demonstrate that there is only one 1-round iterative linear characteristic with three active S-boxes. Since the 1-round linear characteristic is unique, it must be included in any $R$-round ($R \geqslant 12$) linear characteristics of BAKSHEESH with three active S-boxes per round. Finally, we confirm that BAKSHEESH's total number of $R$-round optimal linear characteristics is $3072$ for $R \geqslant 12$. All of these characteristics are generated by employing the 1-round iterative linear characteristic.

In this paper, we investigate whether the privacy mechanism of periodically changing the pseudorandom identities of Bluetooth Low Energy (BLE) beacons is sufficient to ensure privacy. We consider a new natural privacy notion for BLE broadcasting beacons which we call ``Timed-sequence- indistinguishability'' of beacons. This new privacy definition is stronger than the well-known indistinguishability, since it considers not just the advertisements' content, but also the advertisements' broadcasting times which are observable in the physical world. We then prove that beacons with periodically changing pseudorandom identities do not achieve timed-sequence- indistinguishability. We do this by presenting a novel privacy attack against BLE beacons, which we call the ``Timer Manipulation Attack.'' This new time-based privacy attack can be executed by merely inserting or reinserting the beacon's battery at the adversary's chosen time. We performed this attack against an actually deployed beacon. To mitigate the ``Timer Manipulation Attack'' and other attacks associated with periodic signaling, we propose a new countermeasure involving quasi-periodic randomized scheduling of identity changes. We prove that our countermeasure ensures timed-sequence indistinguishability for beacons, thereby enhancing the beacon's privacy. Additionally, we show how to integrate this countermeasure in the attacked system while essentially preserving its feasibility and utility, which is crucial for practical industrial adoption.

In this note we review the technique proposed at ToSC 2018 by Sadeghi et al. for attacks built upon several related-tweakey impossible differential trails. We show that the initial encryption queries are improper and lead the authors to misevaluate a filtering value in the key recovery phase. We identified 4 other papers (from Eurocrypt, DCC, and 2 from ToSC) that follow on the results of Sadeghi et al., and in three of them the flawed technique was reused. We thus present a careful analysis of these types of attacks and give generic complexity formulas similar to the ones proposed by Boura et al. at Asiacrypt 2014. We apply these to the aforementioned papers and provide patched versions of their attacks. The main consequence is an increase in the memory complexity. We show that in many cases (a notable exception being quantum impossible differentials) it is possible to recover the numeric time estimates of the flawed analysis, and in all cases we were able to build a correct attack reaching the same number of rounds.

Elliptic Curve Point Multiplication (ECPM) is a key component of the Elliptic Curve Cryptography (ECC) hierarchy protocol. However, the specific estimation of resources required for this process remains underexplored despite its significance in the cryptanalysis of ECC algorithms, particularly binary ECC in GF (2𝑚). Given the extensive use of ECC algorithms in various security protocols and devices, it is essential to conduct this examination to gain valuable insights into its cryptanalysis, specifically in terms of providing precise resource estimations, which serve as a solid basis for further investigation in solving the Elliptic Curve Discrete Logarithm Problem. Expanding on several significant prior research, in this work, we refer to as ECPM cryptanalysis, we estimate quantum resources, including qubits, gates, and circuit depth, by integrating point addition (PA) and point-doubling (PD) into the ECPM scheme, culminating in a Shor’s algorithm-based binary ECC cryptanalysis circuit. Focusing on optimizing depth, we elaborate on and implement the most efficient PD circuit and incorporate optimized Karatsuba multiplication and FLT-based inversion algorithms for PA and PD operations. Compared to the latest PA-only circuits, our preliminary results showcase significant resource optimization for various ECPM implementations, including single-step ECPM, ECPM with combined or selective PA/PD utilization, and total−step ECPM (2𝑛 PD+2 PA).

Recently, Biryukov et al. presented a new technique for key recovery in differential cryptanalysis, called meet-in-the-filter (MiF). In this work, we develop theoretical and practical aspects of the technique, which helps understanding and simplifies application. In particular, we show bounds on MiF complexity and conditions when the MiF-enhanced attack may reach them. We present a method based on trail counting which allows to estimate filtering strength of involved rounds and perform consequent complexity analysis with pen and paper, compared to the computer-aided approach of the original work. Furthermore, we show how MiF can be combined with plaintext structures for linear key schedules, allowing to increase the number of attacked rounds or to reduce the data complexity. We illustrate our methods on block cipher families CHAM and KATAN and show best-to-date single-key differential attacks for these ciphers.

With the advent of quantum computers, the security of cryptographic primitives, including digital signature schemes, has been compromised. To deal with this issue, some signature schemes have been introduced to resist against these computers. These schemes are known as post-quantum signature schemes. One group of these schemes is based on the hard problems of coding theory, called code-based cryptographic schemes. Several code-based signature schemes are inspired by the McEliece encryption scheme using three non-singular, parity-check, and permutation matrices as the only components of the private keys, and their product as the public key. In this paper, we focus on the analysis of a class of such signature schemes. For this purpose, we first prove that the linear relationships between the columns of the parity-check/generator matrix appear in the public key matrix, and by exploiting this feature we perform a forgery attack on one of the signature schemes of this class as an evidence. The complexity of this attack is of O(n^4).

Evaluation of cryptographic implementations with respect to side-channels has been mandated at high security levels nowadays. Typically, the evaluation involves four stages: detection, modeling, certification and secret recovery. In pursuit of specific goal at each stage, inherently different techniques used to be considered necessary. However, since the recent works of Eurocrypt2022 and Eurocrypt2024, linear regression analysis (LRA) has uniquely become the technique that is well-applied throughout all the stages. In this paper, we concentrate on this silver bullet technique within the field of side-channel. First, we address the fundamental problems of why and how to use LRA. The discussion of nominal and binary nature explains its strong applicability. To sustain effective outcomes, we provide in-depth analyses about the design matrix, regarding the sample distribution of plaintext and the chosen polynomial degree. We summarize ideal conditions that totally avoid multicollinearity problem, and explore the novel evaluator-advantageous property of LRA by means of model diagnosis. Then, we trace the roots where we theoretically elaborate its connections with traditional side-channel techniques, including Correlation Power Analysis (CPA), Distance-of-Means analysis (DoM) and Partition Power Analysis (PPA), in terms of regression coefficients, regression model and coefficient of determination. Finally, we probe into the state-of-the-art combined LRA with the so-called collapse function, demonstrating its relationship with another refined technique, G-DoM. We argue that properly relaxing the definition of bit groups equally satisfies our conclusions. Experimental results are in line with the theory, confirming its correctness.

This paper addresses the problem of factoring composite numbers by introducing a novel approach to represent their prime divisors. We develop a method to efficiently identify smaller divisors based on the difference between the primes involved in forming the composite number. Building on these insights, we propose an algorithm that significantly reduces the computational complexity of factoring, requiring half as many iterations as traditional quadratic residue-based methods. The presented algorithm offers a more efficient solution for factoring composite numbers, with potential applications in fields such as cryptography and computational number theory.

In this paper, we introduce OT-PCA, a novel approach for conducting Plaintext-Checking (PC) oracle based side-channel attacks, specifically designed for Hamming Quasi-Cyclic (HQC). By calling the publicly accessible HQC decoder, we build offline templates that enable efficient extraction of soft information for hundreds of secret positions with just a single PC oracle call. Our method addresses critical challenges in optimizing key-related information extraction, including maximizing decryption output entropy and ensuring error pattern independence, through the use of genetic-style algorithms. Extensive simulations demonstrate that our new attack method significantly reduces the required number of oracle calls, achieving a 2.4-fold decrease for hqc-128 and even greater reductions for hqc-192 and hqc-256 compared to current state-of-the-art methods. Notably, the attack shows strong resilience against inaccuracy in the PC oracle—when the oracle accuracy decreases to 95%, the reduction factor in oracle call requirements increases to 7.6 for hqc-128. Lastly, a real-world evaluation conducted using power analysis on a platform with an ARM Cortex-M4 microcontroller validates the practical applicability and effectiveness of our approach.

Classic McEliece is one of the three code-based candidates in the fourth round of the NIST post-quantum cryptography standardization process in the Key Encapsulation Mechanism category. As such, its decapsulation algorithm is used to recover the session key associated with a ciphertext using the private key. In this article, we propose a new side-channel attack on the syndrome computation in the decapsulation algorithm that recovers the private key, which consists of the private Goppa polynomial $g$ and the permuted support $\mathcal{L}$. The attack relies on both practical aspects and theoretical contributions, namely that the side-channel distinguisher can accurately discriminate elements of the permuted support $\mathcal{L}$, while relying only on a standard noisy Hamming weight leakage assumption and that there exists a cubic-time algorithm that uses this information to recover the private Goppa polynomial $g$. Compared with previous work targeting the Classic McEliece private key, this drastically improves both on the assumptions made in the attacker model and on the overall efficiency of the key-recovery algorithm. We have carried out the attack in practice on a microcontroller target running the reference implementation of Classic McEliece, and make the full attack source code available.

Previous studies on deep-learning-based side-channel attacks (DL-SCAs) have shown that traditional performance evaluation metrics commonly used in DL, like accuracy and F1 score, are not effective in evaluating DL-SCA performance. Therefore, some previous studies have proposed new alternative metrics for evaluating the performance of DL-SCAs. Notably, perceived information (PI) and effective perceived information (EPI) are major metrics based on information theory. While it has been experimentally confirmed that these metrics can give the attack success rate (SR) for DL-SCAs, their theoretical validity remains unclear. In this paper, we propose a new theoretically valid performance evaluation metric called latent perceived information (LPI), which serves as an alternative to the existing metrics. LPI is defined as the mutual information between the output of the feature extractor of a neural network (NN) model and the intermediate value, representing the potential attack performance of the trained model. First, we prove that LPI provides an upper bound on the SR of a DL-SCA by modeling and formulating DL-SCA as a communication channel. Additionally, we clarify the conditions under which PI and EPI theoretically provide an upper bound on the SR from the perspective of LPI. For practical computation of LPI, we present two methods. One utilizes the Kraskov (KSG) estimator, a common mutual information estimator, and the other is based on logistic regression. While the KSG estimator is computationally intensive, it yields accurate LPI values. In contrast, the logistic regression is faster but provides a lower bound for LPI. Through experimental attacks on AES software and hardware implementations with masking countermeasures, we demonstrate that the LPI values estimated by these two methods are significantly similar, indicating the reliability and soundness of our proposed estimation techniques. Furthermore, we show that, by using the logistic regression as a classifier, we can significantly improve the attack performance of the trained model when the difference between the SR upper bound by the LPI and its actual SR is large. This indicates that LPI represents the potential for performance improvement in the trained model. Therefore, our study contributes to optimizing the distinguisher for attack performance using the trained model.

Large language models (LLMs) have recently transformed many industries, enhancing content generation, customer service agents, data analysis and even software generation. These applications are often hosted on remote servers to protect the neural-network model IP; however, this raises concerns about the privacy of input queries. Fully Homomorphic Encryption (FHE), an encryption technique that allows for computations on private data, has been proposed as a solution to the challenge. Nevertheless, due to the increased size of LLMs and the computational overheads of FHE, today's practical FHE LLMs are implemented using a split model approach. Here, a user sends their FHE encrypted data to the server to run an encrypted attention head layer; then the server returns the result of the layer for the user to run the rest of the model locally. By employing this method, the server maintains part of their model IP, and the user still gets to perform private LLM inference. In this work, we evaluate the neural-network model IP protections of single layer split model LLMs, and demonstrate a novel attack vector that makes it easy for a user to extract the neural network model IP from the server, bypassing the claimed protections for encrypted computation. In our analysis, we demonstrate the feasibility of this attack, and discuss potential mitigations.

Key agreement and public key encryption are two elementary cryptographic primitives, suitable for different scenarios. But their differences are still not familiar to some researchers. In this note, we show that the Safkhani et al.'s key agreement scheme [Peer-to-Peer Netw. Appl. 15(3), 1595-1616, 2022] is a public key encryption in disguise. We stress that the ultimate use of key agreement is to establish a shared key for some symmetric key encryption. We also present a simplification of the scheme by removing some repetitive computations. To the best of our knowledge, it is the first time to clarify the fundamental differences between the two primitives. The techniques developed in this note will be helpful for the future works on designing such schemes.

End-to-end encrypted cloud storage offers a way for individuals and organisations to delegate their storage needs to a third-party, while keeping control of their data using cryptographic techniques. We conduct a cryptographic analysis of various products in the ecosystem, showing that many providers fail to provide an adequate level of security. In particular, we provide an in-depth analysis of five end-to-end encrypted cloud storage systems, namely Sync, pCloud, Icedrive, Seafile, and Tresorit, in the setting of a malicious server. These companies cumulatively have over 22 million users and are major providers in the field. We unveil severe cryptographic vulnerabilities in four of them. Our attacks invalidate the marketing claims made by the providers of these systems, showing that a malicious server can, in some cases, inject files in the encrypted storage of users, tamper with file data, and even gain direct access to the content of the files. Many of our attacks affect multiple providers in the same way, revealing common failure patterns in independent cryptographic designs. We conclude by discussing the significance of these patterns beyond the security of the specific providers.

The (parallel) classical black pebbling game is a helpful abstraction which allows us to analyze the resources (time, space, space-time, cumulative space) necessary to evaluate a function $f$ with a static data-dependency graph $G$ on a (parallel) computer. In particular, the parallel black pebbling game has been used as a tool to quantify the (in)security of Data-Independent Memory-Hard Functions (iMHFs). However, the classical black pebbling game is not suitable to analyze the cost of quantum preimage attack. Thus, Blocki et al. (TCC 2022) introduced the parallel reversible pebbling game as a tool to analyze resource requirements for a quantum computer. While there is an extensive line of work analyzing pebbling complexity in the (parallel) black pebbling game, comparatively little is known about the parallel reversible pebbling game. Our first result is a lower bound of $\Omega\left(N^{1+\sqrt{\frac{ 2-o(1)}{\log N}}} \right)$ on the reversible cumulative pebbling cost for a line graph on $N$ nodes. This yields a separation between classical and reversible pebbling costs demonstrating that the reversibility constraint can increase cumulative pebbling costs (and space-time costs) by a multiplicative factor of $N^{(\sqrt 2 + o(1))/\sqrt{\log N}}$ --- the classical pebbling cost (space-time or cumulative) for a line graph is just $\mathcal{O}(N)$. On the positive side, we prove that any classical parallel pebbling can be transformed into a reversible pebbling strategy whilst increasing space-time (resp. cumulative memory) costs by a multiplicative factor of at most $\mathcal{O}\left(N^{\sqrt{\frac{8}{\log N}}}\right)$ (resp. $\mathcal{O}\left(N^{\mathcal{O}(1)/\sqrt[4]{\log N}}\right)$). We also analyze the impact of the reversibility constraint on the cumulative pebbling cost of depth-robust and depth-reducible DAGs exploiting reversibility to improve constant factors in a prior lower bound of Alwen et al. (EUROCRYPT 2017). For depth-reducible DAGs we show that the state-of-the-art recursive pebbling techniques of Alwen et al. (EUROCRYPT 2017) can be converted into a recursive reversible pebbling attack without any asymptotic increases in pebbling costs. Finally, we extend a result of Blocki et al. (ITCS 2020) to show that it is Unique Games hard to approximate the reversible cumulative pebbling cost of a DAG $G$ to within any constant factor.

Lattice cryptography schemes based on the learning with errors (LWE) hardness assumption have been standardized by NIST for use as post-quantum cryptosystems, and by HomomorphicEncryption.org for encrypted compute on sensitive data. Thus, understanding their concrete security is critical. Most work on LWE security focuses on theoretical estimates of attack performance, which is important but may overlook attack nuances arising in real-world implementations. The sole existing concrete benchmarking effort, the Darmstadt Lattice Challenge, does not include benchmarks relevant to the standardized LWE parameter choices - such as small secret and small error distributions, and Ring-LWE (RLWE) and Module-LWE (MLWE) variants. To improve our understanding of concrete LWE security, we provide the first benchmarks for LWE secret recovery on standardized parameters, for small and low-weight (sparse) secrets. We evaluate four LWE attacks in these settings to serve as a baseline: the Search-LWE attacks uSVP, SALSA, and Cool & Cruel, and the Decision-LWE attack: Dual Hybrid Meet-in-the-Middle (MitM). We extend the SALSA and Cool & Cruel attacks in significant ways, and implement and scale up MitM attacks for the first time. For example, we recover hamming weight $9-11$ binomial secrets for KYBER ($\kappa=2$) parameters in $28-36$ hours with SALSA and Cool & Cruel, while we find that MitM can solve Decision-LWE instances for hamming weights up to $4$ in under an hour for Kyber parameters, while uSVP attacks do not recover any secrets after running for more than $1100$ hours. We also compare concrete performance against theoretical estimates. Finally, we open source the code to enable future research.

The Crossbred algorithm is currently the state-of-the-art method for solving overdetermined multivariate polynomial systems over $\mathbb{F}_2$. Since its publication in 2015, several record breaking implementations have been proposed and demonstrate the power of this hybrid approach. Despite these practical results, the complexity of this algorithm and the choice of optimal parameters for it are difficult open questions. In this paper, we prove a bivariate generating series for potentially admissible parameters of the Crossbred algorithm.

The $k$-Tree algorithm [Wagner 02] is a non-trivial algorithm for the average-case $k$-SUM problem that has found widespread use in cryptanalysis. Its input consists of $k$ lists, each containing $n$ integers from a range of size $m$. Wagner's original heuristic analysis suggested that this algorithm succeeds with constant probability if $n \approx m^{1/(\log{k}+1)}$, and that in this case it runs in time $O(kn)$. Subsequent rigorous analysis of the algorithm [Lyubashevsky 05, Shallue 08, Joux-Kippen-Loss 24] has shown that it succeeds with high probability if the input list sizes are significantly larger than this. We present a broader rigorous analysis of the $k$-Tree algorithm, showing upper and lower bounds on its success probability and complexity for any size of the input lists. Our results confirm Wagner's heuristic conclusions, and also give meaningful bounds for a wide range of list sizes that are not covered by existing analyses. We present analytical bounds that are asymptotically tight, as well as an efficient algorithm that computes (provably correct) bounds for a wide range of concrete parameter settings. We also do the same for the $k$-Tree algorithm over $\mathbb{Z}_m$. Finally, we present experimental evaluation of the tightness of our results.

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

Multiple Matrix congruential generators is an important class of pseudorandom number generators. This paper studies the predictability of a class of truncated multiple matrix congruential generators with unknown parameters. Given a few truncated digits of high-order bits or low-order bits output by a multiple matrix congruential generator, we give a method based on lattice reduction to recover the parameters and the initial state of the generator.

We show that the outsourcing algorithm for the case of linear constraints [IEEE Trans. Cloud Comput., 2023, 11(1), 484-498] cannot keep output privacy, due to the simple translation transformation. We also suggest a remedy method by adopting a hybrid transformation which combines the usual translation transformation and resizing transformation so as to protect the output privacy.

Let $(N,e)$ be a public key of the RSA cryptosystem, and $d$ be the corresponding private key. In practice, we usually choose a small $e$ for quick encryption. In this paper, we improve partial private key exposure attacks against RSA with MSBs of $d$ and small $e$. The key idea is that under such a setting we can usually obtain more information about the prime factors of $N$ and then, by solving a univariate modular polynomial equation using Coppersmith's method, $N$ can be factored in polynomial time. Compared to previous results, we reduce the number of the leaked bits in $d$ that are needed to mount the attack by $\log_2 (e)$ bits. For $e=65537$, previous work required an additional enumeration of 17 bits to achieve our new bound, resulting in a $2^{10}$ (or 1,024) x increase in time consumption. Furthermore, our experiments show that for a $1024$-bit modulus $N$, our attack can achieve the theoretical bound on a simple personal computer, which verifies the new method.

Recently, deep learning-based side-channel analysis (DLSCA) has emerged as a serious threat against cryptographic implementations. These methods can efficiently break implementations protected with various countermeasures while needing limited manual intervention. To effectively protect implementation, it is therefore crucial to be able to interpret \textbf{how} these models are defeating countermeasures. Several works have attempted to gain a better understanding of the mechanics of these models. However, a fine-grained description remains elusive. To help tackle this challenge, we propose using Kolmogorov-Arnold Networks (KANs). These neural networks were recently introduced and showed competitive performance to multilayer perceptrons (MLPs) on small-scale tasks while being easier to interpret. In this work, we show that KANs are well suited to SCA, performing similarly to MLPs across both simulated and real-world traces. Furthermore, we find specific strategies that the trained models learn for combining mask shares and are able to measure what points in the trace are relevant.

The Fiat-Shamir transformation is a key technique for removing interactivity from cryptographic proof systems in real-world applications. In this work, we discuss five types of Fiat-Shamir-related protocol design errors and illustrate them with concrete examples mainly taken from real-life applications. We discuss countermeasures for such vulnerabilities.

Deep neural networks have proven effective for second-order profiling side-channel attacks, even in a black-box setting with no prior knowledge of masks and implementation details. While such attacks have been successful, no explanations were provided for understanding why a variety of deep neural networks can (or cannot) learn high-order leakages and what the limitations are. In other words, we lack the explainability on neural network layers combining (or not) unknown and random secret shares, which is a necessary step to defeat, e.g., Boolean masking countermeasures. In this paper, we use information-theoretic metrics to explain the internal activities of deep neural network layers. We propose a novel methodology for the explainability of deep learning-based profiling side-channel analysis (denoted ExDL-SCA) to understand the processing of secret masks. Inspired by the Information Bottleneck theory, our explainability methodology uses perceived information to explain and detect the different phenomena that occur in deep neural networks, such as fitting, compression, and generalization. We provide experimental results on masked AES datasets showing what relevant features deep neural networks use, and where in the networks relevant features are learned and irrelevant features are compressed. Using our method, evaluators can determine what secret masks are being exploited by the network, which allows for more detailed feedback on the implementations. This paper opens new perspectives for understanding the role of different neural network layers in profiling side-channel attacks.

We use the HHL algorithm to retrieve a quantum state holding the algebraic normal formal of a Boolean function. Unlike the standard HHL applications, we do not describe the cipher as an exponentially big system of equations. Rather, we perform a set of small matrix inversions which corresponds to the Boolean Möbius transform. This creates a superposition holding information about the ANF in the form: $\ket{\mathcal{A}_{f}} =\frac{1}{C} \sum_{I=0}^{2^n-1} c_I \ket{I}$, where $c_I$ is the coefficient of the ANF and $C$ is a scaling factor. The procedure has a time complexity of $\mathcal{O}(n)$ for a Boolean function with $n$ bit input. We also propose two approaches how some information about the ANF can be extracted from such a state.

NIST released the draft standard for ML-KEM, and we can expect its widespread use in the embedded world in the near future. Several side-channel attacks have been proposed, and one line of research has focused on attacks against the comparison step of the FO-transform. A work published at TCHES 2022 stressed the need for secure higher-order masked comparisons beyond the $t$-probing model and proposed a higher-order masked comparison method. Subsequently, D'Anvers, Van Beirendonck, and Verbauwhede improved upon the performance of several previous proposals; their higher-order masked algorithm currently achieves the highest performance for masked comparisons. In this work, we show that while this proposal is secure in the $t$-probing model, its security in practice is questionable. We first propose an approximate template attack that requires only a small number of traces for profiling and has an exceptionally high noise tolerance. We demonstrate that, without knowledge of the targeted values, a vertical analysis of the distribution of certain points of interest can replace the profiling phase. Finally, we explain how a decryption failure oracle may be constructed from a single trace. We prove that these attacks are not affected by higher masking orders for noise levels that by far prevent previous profiled attacks on ML-KEM. Further, we provide simulations showing that even under extreme noise levels, the attacks are not prevented by realistic masking orders. Additionally, we carry out the attacks on multiple physical devices to stress the practicality of our attack. We discuss the underlying causes for our attack, demonstrate the difficulty of securing the FO-transform in ML-KEM, draw conclusions about the (in-)sufficiency of $t$-probing security in this context, and highlight an open gap in securing ML-KEM on embedded devices.

This paper questions the side-channel security of central reduction technique, which is widely adapted in efficient implementations of Lattice-Based Cryptography (LBC). We show that the central reduction leads to a vulnerability by creating a strong dependency between the power consumption and the sign of sensitive intermediate values. We exploit this dependency by introducing the novel absolute value prediction function, which can be employed in higher-order non-profiled multi-query Side-Channel Analysis (SCA) attacks. Our results reveal that – compared to classical reduction algorithms – employing the central reduction scheme leads to a two-orders-of-magnitude decrease in the number of required SCA measurements to exploit secrets of masked implementations. We particularly show that our approach is valid for the prime moduli employed by Kyber and Dilithium, the lattice-based post-quantum algorithms selected by NIST. We practically evaluate our introduced approach by performing second-order non-profiled attacks against an open-source masked implementation of Kyber on an ARM Cortex-M4 micro-processor. In our experiments, we revealed the full secret key of the aforementioned masked implementation with only 250 power traces without any forms of profiling or choosing the ciphertexts.

We analyse HWQCS, a code based signature scheme presented at ICISC 2023, which uses quasi-cyclic low density parity check codes (QC-LDPC). The scheme introduces high Hamming weight errors and signs each message using a fresh ephemeral secret key rather than using only one secret key, so to avoid known attacks on QC-LDPC signature schemes. In this paper, we show that the signatures of HWQCS leak substantial information concerning the ephemeral keys and formally describe this behaviour. Furthermore, we show that for each security level, we can exploit the leakage to efficiently reconstruct partial secret data from very few signatures, and finally mount a universal forgery attack.

Encrypted Search Algorithms (ESAs) are a technique to encrypt data while the user can still search over it. ESAs can protect privacy and ensure security of sensitive data stored on a remote storage. Originally, ESAs were used in the context of documents that consist of keywords. The user encrypts the documents, sends them to a remote server and is still able to search for keywords, without exposing information about the plaintext. The idea of ESAs has also been applied to relational databases, where queries (similar to SQL statements) can be privately executed on an encrypted database.But just as traditional schemes for Keyword-ESAs, also Relational-ESAs have the drawback of exposing some information, called leakage. Leakage attacks have been proposed in the literature that use this information together with auxiliary information to learn details about the plaintext. However, these leakage attacks have overwhelmingly been designed for and applied to Keyword-ESAs and not Relational-ESAs. In this work, we review the suitability of major leakage attacks against ESAs in the relational setting by adapting them accordingly. We perform extensive re-evaluations of the attacks on various relational datasets with different properties. Our evaluations show that major attacks can work against Relational-ESAs in the known-data setting. However, the attack performance differs between datasets, exploited patterns, and attacks.

In CRYPTO'19, Gohr introduced a novel cryptanalysis method by developing a differential-neural distinguisher using neural networks as the underlying distinguisher. He effectively integrated this distinguisher with classical differentials, facilitating a 12-round key recovery attack on Speck32/64 (from a total of 22 rounds). Bao et al. refined the concept of neutral bits, enabling key recovery attacks up to 13 rounds for Speck32/64 and 16 rounds (from a total of 32) for Simon32/64. Our primary objective is to enhance the capabilities of differential-neural distinguishers by applying more deep-learning techniques, focusing on handling more rounds and improving accuracy. Inspired by the Inception Block in GoogLeNet, we adopted a design that uses multiple parallel convolutional layers with varying kernel sizes before the residual block to capture multi-dimensional information. Additionally, we expanded the convolutional kernels in the residual blocks, thereby enlarging the network's receptive field. In the case of Speck32/64, our efforts yield accuracy improvements in rounds 6, 7, and 8, enabling the successful training of a 9-round differential-neural distinguisher. As for Simon32/64, we developed a differential-neural distinguisher capable of effectively handling 12 rounds while achieving noteworthy accuracy enhancements in rounds 9, 10, and 11. Additionally, we utilized neutral bits to ensure the required data distribution for launching a successful key recovery attack when using multiple-ciphertext pairs as input for the neural network. Meanwhile, we redefined the formula for time complexity based on the differences in prediction speeds of the distinguisher between a single-core CPU and a GPU. Combining these various advancements allows us to considerably reduce the time and data complexity of key recovery attacks on 13-round Speck32/64. Furthermore, we used knowledge distillation techniques to reduce the model size, thereby accelerating the distinguisher's prediction speed and reducing the time complexity. In particular, we achieved a successful 14-round key recovery attack by exhaustively guessing a 1-round subkey, marking a significant milestone in differential-neural cryptanalysis. For Simon32/64, we accomplished a groundbreaking 17-round key recovery attack for the first time and reduced the time complexity of the 16-round key recovery attack.

In this paper, we investigate the rough order assumption (\(RO_C\)) introduced by Braun, Damgård, and Orlandi at CRYPTO 23, which posits that class groups of imaginary quadratic fields with no small prime factors in their order are computationally indistinguishable from general class groups. We present a novel attack that challenges the validity of this assumption by leveraging properties of Mordell curves over the rational numbers. Specifically, we demonstrate that if the rank of the Mordell curve \( E_{-16D} \) is at least 2, it contradicts the rough order assumption. Our attack deterministically breaks the \(RO_C\) assumption for discriminants of a special form, assuming the parity conjecture holds and certain conditions are met. Additionally, for both special and generic cases, our results suggest that the presence of nontrivial 3-torsion elements in class groups can undermine the \(RO_C\) assumption. Although our findings are concrete for specific cases, the generic scenario relies on heuristic arguments related to the Birch and Swinnerton-Dyer (BSD) conjecture, a significant and widely believed conjecture in number theory. Attacks against 2-torsion elements in class groups are already well known, but our work introduces a distinct approach targeting 3-torsion elements. These attacks are fundamentally different in nature, and both have relatively straightforward countermeasures, though they do not generalize to higher torsions. While these results do not entirely invalidate the \(RO_C\) assumption, they highlight the need for further exploration of its underlying assumptions, especially in the context of specific torsion structures within class groups.

At Eurocrypt'24, Mureau et al. formally defined the Lattice Isomorphism Problem for module lattices (module-LIP) in a number field $\mathbb{K}$, and proposed a heuristic randomized algorithm solving module-LIP for modules of rank 2 in $\mathbb{K}^2$ with a totally real number field $\mathbb{K}$, which runs in classical polynomial time for a large class of modules and a large class of totally real number field under some reasonable number theoretic assumptions. In this paper, by introducing a (pseudo) symplectic automorphism of the module, we successfully reduce the problem of solving module-LIP over CM number field to the problem of finding certain symplectic automorphism. Furthermore, we show that a weak (pseudo) symplectic automorphism can be computed efficiently, which immediately turns out to be the desired automorphism when the module is in a totally real number field. This directly results in a provable deterministic polynomial-time algorithm solving module-LIP for rank-2 modules in $\mathbb{K}^2$ where $\mathbb{K}$ is a totally real number field, without any assumptions or restrictions on the modules and the totally real number fields. Moreover, the weak symplectic automorphism can also be utilized to invalidate the omSVP assumption employed in HAWK's forgery security analysis, although it does not yield any actual attacks against HAWK itself.

Recently, Geraud-Stewart and Naccache proposed two trapdoors based on matrix products. In this paper, we answer the call for cryptanalysis. We explore how using the trace and determinant of a matrix can be used to attack their constructions. We fully break their first construction in a polynomial-time attack. We show an information leak in the second construction using characteristic polynomials, and provide an attack using traces that decreases the bit security by about half.

Peer-to-peer communication systems can provide many functions, including anonymized routing of network traffic, massive parallel computing environments, and distributed storage. Anonymity refers to the state of being completely nameless, with no attached identifiers. Pseudonymity involves the use of a fictitious name that can be consistently linked to a particular user, though not necessarily to the real identity. Both provide a layer of privacy, shielding the user's true identity from public view. But we find their significations are often misunderstood. In this note, we clarify the differences between anonymity and pseudonymity. We also find the Zhong et al.'s key agreement scheme [IEEE TCC, 2022, 10(3), 1592-1603] fails to keep anonymity, not as claimed.

We present an efficient quantum algorithm for solving the semidirect discrete logarithm problem (SDLP) in any finite group. The believed hardness of the semidirect discrete logarithm problem underlies more than a decade of works constructing candidate post-quantum cryptographic algorithms from nonabelian groups. We use a series of reduction results to show that it suffices to consider SDLP in finite simple groups. We then apply the celebrated Classification of Finite Simple Groups to consider each family. The infinite families of finite simple groups admit, in a fairly general setting, linear algebraic attacks providing a reduction to the classical discrete logarithm problem. For the sporadic simple groups, we show that their inherent properties render them unsuitable for cryptographically hard SDLP instances, which we illustrate via a Baby-Step Giant-Step style attack against SDLP in the Monster Group. Our quantum SDLP algorithm is fully constructive, up to the compu- tation of maximal normal subgroups, for all but three remaining cases that appear to be gaps in the literature on constructive recognition of groups; for these cases SDLP is no harder than finding a linear representation. We conclude that SDLP is not a suitable post-quantum hardness assumption for any choice of finite group.

The impossible differential (ID) attack is one of the most important cryptanalytic techniques for block ciphers. There are two phases to finding an ID attack: searching for the distinguisher and building a key recovery upon it. Previous works only focused on automated distinguisher discovery, leaving key recovery as a manual post-processing task, which may lead to a suboptimal final complexity. At EUROCRYPT~2023, Hadipour et al. introduced a unified constraint programming (CP) approach based on satisfiability for finding optimal complete ID attacks in strongly aligned ciphers. While this approach was extended to weakly-aligned designs like PRESENT at ToSC~2024, its application to ARX and AndRX ciphers remained as future work. Moreover, this method only exploited ID distinguishers with direct contradictions at the junction of two deterministic transitions. In contrast, some ID distinguishers, particularly for ARX and AndRX designs, may not be detectable by checking only the existence of direct contradictions. This paper fills these gaps by extending Hadipour et al.'s method to handle indirect contradictions and adapting it for ARX and AndRX designs. We also present a similar method for identifying zero-correlation (ZC) distinguishers. Moreover, we extend our new model for finding ID distinguishers to a unified optimization problem that includes both the distinguisher and the key recovery for AndRX designs. Our method improves ID attacks and introduces new distinguishers for several ciphers, such as SIMON, SPECK, Simeck, ChaCha, Chaskey, LEA, and SipHash. For example, we achieve a one-round improvement in the ID attacks against SIMON-64-96, SIMON-64-128, SIMON-128-128, SIMON-128-256 and a two-round improvement in the ID attacks against SIMON-128-192. These results significantly contribute to our understanding of the effectiveness of automated tools in the cryptanalysis of different design paradigms.

The Hidden Number Problem (HNP) has found extensive applications in side-channel attacks against cryptographic schemes, such as ECDSA and Diffie-Hellman. There are two primary algorithmic approaches to solving the HNP: lattice-based attacks and Fourier analysis-based attacks. Lattice-based attacks exhibit better efficiency and require fewer samples when sufficiently long substrings of the nonces are known. However, they face significant challenges when only a small fraction of the nonce is leaked, such as 1-bit leakage, and their performance degrades in the presence of errors. In this paper, we address an open question by introducing an algorithmic tradeoff that significantly bridges the gap between these two approaches. By introducing a parameter $x$ to modify Albrecht and Heninger's lattice, the lattice dimension is reduced by approximately $(\log_2{x})/ l$, where $l$ represents the number of leaked bits. We present a series of new methods, including the interval reduction algorithm, several predicates, and the pre-screening technique. Furthermore, we extend our algorithms to solve the HNP with erroneous input. Our attack outperforms existing state-of-the-art lattice-based attacks against ECDSA. We break several records including 1-bit and less than 1-bit leakage on a 160-bit curve, while the best previous lattice-based attack for 1-bit leakage was conducted only on a 112-bit curve.

We present a key recovery attack on SQIsign2D-East that reduces its security level from $\lambda$ to $\lambda/2$. We exploit the fact that each signature leaks a Legendre symbol modulo the secret degree of the private key isogeny. About $\lambda/2$ signatures are enough for these Legendre symbols to fully determine the secret degree, which can then be recovered by exhaustive search over a set of size $O(2^{\lambda/2})$. Once the degree is known, the private key isogeny itself can be found, again by exhaustive search, in time $\tilde{O}(2^{\lambda/2})$. We also present a new version of the protocol which does not leak any such information about the private key and show that our modified protocol is more efficient than the original one. Finally, we give a security analysis as well as a new proof of security.

Post-quantum cryptographic (PQC) algorithms, especially those based on the learning with errors (LWE) problem, have been subjected to several physical attacks in the recent past. Although the attacks broadly belong to two classes -- passive side-channel attacks and active fault attacks, the attack strategies vary significantly due to the inherent complexities of such algorithms. Exploring further attack surfaces is, therefore, an important step for eventually securing the deployment of these algorithms. Also, it is important to test the robustness of the already proposed countermeasures in this regard. In this work, we propose a new fault attack on side-channel secure masked implementation of LWE-based key-encapsulation mechanisms (KEMs) exploiting fault propagation. The attack typically originates due to an algorithmic modification widely used to enable masking, namely the Arithmetic-to-Boolean ($\mathtt{A2B}$) conversion. We exploit the data dependency of the adder carry chain in $\mathtt{A2B}$ and extract sensitive information, albeit masking (of arbitrary order) being present. As a practical demonstration of the exploitability of this information leakage, we show key recovery attacks of Kyber, although the leakage also exists for other schemes like Saber. The attack on Kyber targets the decapsulation module and utilizes Belief Propagation (BP) for key recovery. To the best of our knowledge, it is the first attack exploiting an algorithmic component introduced to ease masking rather than only exploiting the randomness introduced by masking to obtain desired faults (as done by Delvaux). Finally, we performed both simulated and electromagnetic (EM) fault-based practical validation of the attack for an open-source first-order secure Kyber implementation running on an STM32 platform.

The Peregrine signature scheme is one of the candidates in the ongoing Korean post-quantum cryptography competition. It is proposed as a high-speed variant of Falcon, which is a hash-and-sign signature scheme over NTRU lattices and one of the schemes selected by NIST for standardization. To this end, Peregrine replaces the lattice Gaussian sampler in the Falcon signing procedure with a new sampler based on the centered binomial distribution. While this modification offers significant advantages in terms of efficiency and implementation, it does not come with a provable guarantee that signatures do not leak information about the signing key. Unfortunately, lattice based signature schemes in the hash-and-sign paradigm that lack such a guarantee (such as GGH, NTRUSign or DRS) have generally proved insecure. In this paper, we show that Peregrine is no exception, by demonstrating a practical key recovery attack against it. We observe that the distribution of Peregrine signatures is a hidden transformation of some public distribution and still leaks information about the signing key. By adapting the parallelepiped-learning technique of Nguyen and Regev (Eurocrypt 2006), we show that the signing key can be recovered from a relatively small number of signatures. The learning technique alone yields an approximate version of the key, from which we can recover the exact key using a decoding technique due to Thomas Prest (PKC 2023). For the reference implementation (resp. the official specification version) of Peregrine-512, we fully recover the secret key with good probability in a few hours given around 25,000 (resp. 11 million) signature samples.

Among the various threats to secure ICs, many are semi-invasive in the sense that their application requires the removal of the package to gain access to either the front or back of the target IC. Despite this stringent application requirements, little attention is paid to embedded techniques aiming at checking the package's integrity. This paper explores the feasibility of verifying the package integrity of microcontrollers by examining their thermal dissipation capability.

This paper presents an optimization of the memory cost of the quantum Information Set Decoding (ISD) algorithm proposed by Bernstein (PQCrypto 2010), obtained by combining Prange's ISD with Grover's quantum search. When the code has constant rate and length $n$, this algorithm essentially performs a quantum search which, at each iteration, solves a linear system of dimension $\mathcal{O}(n)$. The typical code lengths used in post-quantum public-key cryptosystems range from $10^3$ to $10^5$. Gaussian elimination, which was used in previous works, needs $\mathcal{O}(n^2)$ space to represent the matrix, resulting in millions or billions of (logical) qubits for these schemes. In this paper, we propose instead to use the algorithm for sparse matrix inversion of Wiedemann (IEEE Trans. inf. theory 1986). The interest of Wiedemann's method is that one relies only on the implementation of a matrix-vector product, where the matrix can be represented in an implicit way. This is the case here. We give two main trade-offs, which we have fully implemented, tested on small instances, and benchmarked for larger instances. The first one is a quantum circuit using $\mathcal{O}(n)$ qubits, $\mathcal{O}(n^3)$ Toffoli gates like Gaussian elimination, and depth $\mathcal{O}(n^2 \log n)$. The second one is a quantum circuit using $\mathcal{O}(n \log^2 n)$ qubits, $\mathcal{O}(n^3)$ gates in total but only $\mathcal{O}( n^2 \log^2 n)$ Toffoli gates, which relies on a different representation of the search space. As an example, for the smallest Classic McEliece parameters we estimate that the Quantum Prange's algorithm can run with 18098 qubits, while previous works would have required at least half a million qubits.

The Li et al.'s scheme [Computer Communications, 186 (2022), 110-120)] uses XOR operation to realize the private transmission of sensitive information, under the assumption that if only one parameter in the expression $ a= b\oplus c $ is known, an adversary cannot retrieve the other two. The assumption neglects that the operands $b$ and $c$ must be of the same bit-length, which leads to the exposure of a substring in the longer operand. The scheme wrongly treats timestamps as random strings to encrypt a confidential parameter. These misuses result in the loss of sensor node's anonymity, the loss of user anonymity and untraceability, insecurity against off-line password guessing attack, and insecurity against impersonation attack. The analysis techniques developed in this note is helpful for the future works on designing such schemes.

The machine learning problem of extracting neural network parameters has been proposed for nearly three decades. Functionally equivalent extraction is a crucial goal for research on this problem. When the adversary has access to the raw output of neural networks, various attacks, including those presented at CRYPTO 2020 and EUROCRYPT 2024, have successfully achieved this goal. However, this goal is not achieved when neural networks operate under a hard-label setting where the raw output is inaccessible. In this paper, we propose the first attack that theoretically achieves functionally equivalent extraction under the hard-label setting, which applies to ReLU neural networks. The effectiveness of our attack is validated through practical experiments on a wide range of ReLU neural networks, including neural networks trained on two real benchmarking datasets (MNIST, CIFAR10) widely used in computer vision. For a neural network consisting of $10^5$ parameters, our attack only requires several hours on a single core.

The see-in-the-middle (SITM) attack combines differential cryptanalysis and the ability to observe differential patterns in the side-channel leakage traces to reveal the secret key of SPN-based ciphers. While SITM presents a fresh perspective to side-channel analysis and allows attacks on deeper cipher rounds, there are practical difficulties that come with this method. First, one must realize a visual inspection of millions of power traces. Second, there is a strong requirement to reduce noise to a minimum, achieved by averaging over 1000 traces in the original work, to see the patterns. Third, the presence of a jitter-based countermeasure greatly affects pattern identification, making the visual inspection infeasible. In this paper we aim to tackle these difficulties by using a machine learning approach denoted as DL-SITM (deep learning SITM). The fundamental idea of our approach is that, while a collision obscured by noise is imperceptible in a manual inspection, a powerful deep learning model can identify it, even when a jitter-based countermeasure is in place. As we show with a practical experiment, the proposed DL-SITM approach can distinguish the two valid differentials from over 4M differential traces with only six false positives. Extrapolating from the parameters of this experiment, we get a rough estimate of $2^{43}$ key candidates for the post-processing step of our attack, which places it easily in the practical range. At the same time, we show that even with a jitter countermeasure shifting the execution by $\pm15$ samples, the testing f1-score stays at a relatively high (0.974).

Recently, there has been a lot of interest in improving the understanding of the practical hardness of the 3-Tensor Isomorphism (3-TI) problem, which, given two 3-tensors, asks for an isometry between the two. The current state-of-the-art for solving this problem is the algebraic algorithm of Ran et al. '23 and the graph-theoretic algorithm of Narayanan et al. '24 that have both slightly reduced the security of the signature schemes MEDS and ALTEQ, based on variants of the 3-TI problem (Matrix Code Equivalence (MCE) and Alternating Trilinear Form Equivalence (ATFE) respectively). In this paper, we propose a new combined technique for solving the 3-TI problem. Our algorithm, as typically done in graph-based algorithms, looks for an invariant in the graphs of the isomorphic tensors that can be used to recover the secret isometry. However, contrary to usual combinatorial approaches, our approach is purely algebraic. We model the invariant as a system of non-linear equations and solve it. Using this modelling we are able to find very rare invariant objects in the graphs of the tensors — cycles of length 3 (triangles) — that exist with probability approximately $1/q$. For solving the system of non-linear equations we use Gröbner-basis techniques adapted to tri-graded polynomial rings. We analyze the algorithm theoretically, and we provide lower and upper bounds on its complexity. We further provide experimental support for our complexity claims. Finally, we describe two dedicated versions of our algorithm tailored to the specifics of the MCE and the ATFE problems. The implications of our algorithm are improved cryptanalysis of both MEDS and ALTEQ for the cases when a triangle exists, i.e. in approximately $1/q$ of the cases. While for MEDS, we only marginally reduce the security compared to previous work, for ALTEQ our results are much more significant with at least 60 bits improvement compared to previous work for all security levels. For Level I parameters, our attack is practical, and we are able to recover the secret key in only 1501 seconds. The code is available for testing and verification of our results.

The area of modern zero-knowledge proof systems has seen a significant rise in popularity over the last couple of years, with new techniques and optimized constructions emerging on a regular basis. As the field matures, the aspect of implementation attacks becomes more relevant, however side-channel attacks on zero-knowledge proof systems have seen surprisingly little treatment so far. In this paper we give an overview of potential attack vectors and show that some of the underlying finite field libraries, and implementations of heavily used components like hash functions, are vulnerable w.r.t. cache attacks on CPUs. On the positive side, we demonstrate that the computational overhead to protect against these attacks is relatively small.

The success of deep learning across a variety of applications, including inference on edge devices, has led to increased concerns about the privacy of users’ data and deep learning models. Secure multiparty computation allows parties to remedy this concern, resulting in a growth in the number of such proposals and improvements in their efficiency. The majority of secure inference protocols relying on multiparty computation assume that the client does not deviate from the protocol and passively attempts to extract information. Yet clients, driven by different incentives, can act maliciously to actively deviate from the protocol and disclose the deep learning model owner’s private information. Interestingly, faults are well understood in multiparty computation-related literature, although fault attacks have not been explored. Our paper introduces the very first fault attack against secure inference implementations relying on garbled circuits as a prime example of multiparty computation schemes. In this regard, laser fault injection coupled with a model-extraction attack is successfully mounted against existing solutions that have been assumed to be secure against active attacks. Notably, the number of queries required for the attack is equal to that of the best model extraction attack mounted against the secure inference engines under the semi-honest scenario.