International Association
for Cryptologic Research

During the last decade, there has been a stunning progress in the domain of AI with adoption in both safety-critical and security-critical applications. A key requirement for this is highly trained Machine Learning (ML) models, which are valuable Intellectual Property (IP) of the respective organizations. Naturally, these models have become targets for model recovery attacks through side-channel leakage. However, majority of the attacks reported in literature are either on simple embedded devices or assume a custom Vivado HLS based FPGA accelerator. On the other hand, for commercial neural network accelerators, such as Google TPU, Intel Compute Stick and NVDLA, there are relatively fewer successful attacks. Focussing on that direction, in this work, we study the vulnerabilities of commercial open-source accelerator NVDLA and present the first successful model recovery attack. For this purpose, we use power and timing side-channel leakage information from Convolutional Neural Network (CNN) models to train CNN based attack models. Utilizing these attack models, we demonstrate that even with a highly pipelined architecture, multiple parallel execution in the accelerator along with Linux OS running tasks in the background, recovery of number of layers, kernel sizes, output neurons and distinguishing different layers, is possible with very high accuracy. Our solution is fully automated, and portable to other hardware neural networks, thus presenting a greater threat towards IP protection.

As a typical representative of the public key cryptosystem, RSA has attracted a great deal of cryptanalysis since its invention, among which a famous attack is the small private exponent attack. It is well-known that the best theoretical upper bound for the private exponent d that can be attacked is d ≤ N^0.292 , where N is a RSA modulus. However, this bound may not be achieved in practical attacks since the lattice constructed by Coppersmith method may have a large enough dimension and the lattice-based reduction algorithms cannot work so well in both efficiency and quality. In this paper, we propose a new practical attack based on the binary search for the most significant bits (MSBs) of prime divisors of N and the Herrmann-May’s attack in 2010. The idea of binary search is inspired by the discovery of phenomena called “multivalued-continuous phenomena”, which can significantly accelerate our attack. Together with several carefully selected parameters according to our exact and effective numerical estimations, we can improve the upper bound of d that can be practically achieved. We believe our method can provide some inspiration to practical attacks on RSA with mainstream-size moduli.

We propose how to homomorphically evaluate arbitrary univariate and bivariate integer functions such as division. A prior work proposed by Okada et al. (WISTP'18) uses polynomial evaluations such that the scheme is still compatible with the SIMD operations in BFV and BGV, and is implemented with the input domain size $\mathbb{Z}_{257}$. However, the scheme of Okada et al. requires the quadratic number of plaintext-ciphertext multiplications and ciphertext-ciphertext additions in the input domain size, and although these operations are more lightweight than the ciphertext-ciphertext multiplication, the quadratic complexity makes handling larger inputs quite inefficient. In this work, first we improve the prior work and also propose a new approach that exploits the packing method to handle the larger input domain size instead of enabling the SIMD operation, thus making it possible to work with the larger input domain size, e.g., $\mathbb{Z}_{2^{15}}$ in a reasonably efficient way. In addition, we show how to slightly extend the input domain size to $\mathbb{Z}_{2^{16}}$ with a relatively moderate overhead. Further we show another approach to handling the larger input domain size by using two ciphertexts to encrypt one integer plaintext and applying our techniques for uni/bivariate function evaluation. We implement the prior work of Okada et al., our improved scheme of Okada et al., and our new scheme in PALISADE with the input domain size $\mathbb{Z}_{2^{15}}$, and confirm that the estimated run-times of the prior work and our improved scheme of the prior work are still about 117 days and 59 days respectively while our new scheme can be computed in 307 seconds.

This article presents the application of homomorphic authenticators, replication encodings to be precise, to multigroup fully homomorphic encryption schemes. Following the works of Gennaro and Wichs on homomorphic authenticators in combination with the work of multigroup schemes by Kwak et al. we present a verifiable solution for a fully homomorphic primitive that includes the multikey, multiparty and single user cases. Furthermore, we propose a line of research for the future with constrained-resource scenarios.

This work investigates zero-knowledge protocols in subverted RSA groups where the prover can choose the modulus and where the verifier does not know the group order. We introduce a novel technique for extracting the witness from a general homomorphism over a group of unknown order that does not require parallel repetitions. We present a NIZK range proof for general homomorphisms such as Paillier encryptions in the designated verifier model that works under a subverted setup. The key ingredient of our proof is a constant sized NIZK proof of knowledge for a plaintext. Security is proven in the ROM assuming an IND-CPA additively homomorphic encryption scheme. The verifier's public key is reusable, can be maliciously generated and is linear in the number of proofs to be verified.

Long-term security, a variant of Universally Composable (UC) security introduced by Müller-Quade and Unruh (JoC ’10), allows to analyze the security of protocols in a setting where all hardness assumptions no longer hold after the protocol execution has finished. Such a strict notion is highly desirable when properties such as input privacy need to be guaranteed for a long time, e.g. zero-knowledge proofs for secure electronic voting. Strong impossibility results rule out so-called long-term-revealing setups, e.g. a common reference string (CRS), to achieve long-term security, with known constructions for long-term security requiring hardware assumptions, e.g. signature cards. We circumvent these impossibility results by making use of new techniques, allowing rewinding-based simulation in a way that universal composability is possible. The new techniques allow us to construct a long-term-secure composable commitment scheme in the CRS-hybrid model, which is provably impossible in the notion of Müller-Quade and Unruh. We base our construction on a statistically hiding commitment scheme in the CRS-hybrid model with CCA-like properties. To provide a CCA oracle, we cannot rely on superpolynomial extraction techniques, as statistically hiding commitments do not define a unique value. Thus, we extract the value committed to via rewinding. However, even a CCA “rewinding oracle” without additional properties may be useless, as extracting a malicious committer could require to rewind other protocols the committer participates in. If this is e.g. a reduction, this clearly is forbidden. Fortunately, we can establish the well-known and important property of k-robust extractability, which guarantees that extraction is possible without rewinding k-round protocols the malicious committer participates in. While establishing this property for statistically binding commitment schemes is already non-trivial, it is even more complicated for statistically hiding ones. We then incorporate rewinding-based commitment extraction into the UC framework via a helper in analogy to Canetti, Lin and Pass (FOCS 2010), allowing both adversary and environment to extract statistically hiding commitments. Despite the rewinding, our variant of long-term security is universally composable. Our new framework provides the first setting in which a commitment scheme that is both statistically hiding and composable can be constructed from standard polynomial-time hardness assumptions and a CRS only. Unfortunately, we can prove that our setting does not admit long-term-secure oblivious transfer (and thus general two-party computations). Still, our long-term-secure commitment scheme suffices for natural applications, such as long-term secure and composable (commit-and-prove) zero-knowledge arguments of knowledge.

Current solutions to quantum vulnerabilities of widely used cryptographic schemes involve migrating users to post-quantum schemes before quantum attacks become feasible. This work deals with protecting quantum procrastinators: users that failed to migrate to post-quantum cryptography in time. To address this problem in the context of digital signatures, we introduce a technique called signature lifting, that allows us to lift a deployed pre-quantum signature scheme satisfying a certain property to a post-quantum signature scheme that uses the same keys. Informally, the said property is that a post-quantum one-way function is used "somewhere along the way" to derive the public-key from the secret-key. Our constructions of signature lifting relies heavily on the post-quantum digital signature scheme Picnic (Chase et al., CCS'17). Our main case-study is cryptocurrencies, where this property holds in two scenarios: when the public-key is generated via a key-derivation function or when the public-key hash is posted instead of the public-key itself. We propose a modification, based on signature lifting, that can be applied in many cryptocurrencies for securely spending pre-quantum coins in presence of quantum adversaries. Our construction improves upon existing constructions in two major ways: it is not limited to pre-quantum coins whose ECDSA public-key has been kept secret (and in particular, it handles all coins that are stored in addresses generated by HD wallets), and it does not require access to post-quantum coins or using side payments to pay for posting the transaction.

We study authenticated encryption (AE) modes dedicated to very short messages, which are crucial for Internet-of-things applications. Since the existing general-purpose AE modes need at least three block cipher calls for non-empty messages, we explore the design space for AE modes that use at most two calls. We proposed a family of AE modes, dubbed Manx, that work when the total input length is less than $2n$ bits, using an $n$-bit block cipher. Notably, the second construction of Manx can encrypt almost n-bit plaintext and saves one or two block cipher calls from the standard modes, such as GCM or OCB, keeping the comparable provable security. We also present benchmarks on popular 8/32-bit microprocessors using AES. Our results show the clear advantage of Manx over the previous modes for such short messages.

Digital signatures ensure legitimate access through identity authentication. It is also used to build blocks in blockchains and to authenticate transactions. The Courtois-Finiasz-Sendrier (CFS) digital signature is a well-known code-based digital signature scheme based on the Niederreiter cryptosystem. The CFS signature, however, is not widely used due to the long processing time required by its signing algorithm. Most code-based digital signature schemes are based on Niederreiter. The present paper proposes a new code-based digital signature based on the McEliece cryptosystem. The proposed McEliece code-based scheme also gives less complexity and a higher success rate. The scheme provides an efficient code-based algorithm to sign a document in a shorter processing time. The scheme is also secure against public key structural attacks. Key generation, signing, and verification algorithms are presented. The key generation algorithm constructs three-tuple public keys using a dual inverse matrix. The proposed signing scheme is the first code-based digital signature based on McEliece with the lower processing time required to construct a valid digital signature. The proposed signing algorithm also constructs smaller signatures. In addition, the verification algorithm checks the integrity value to avoid any forgery before final verification.

Machine learning (ML) has seen a strong rise in popularity in recent years and has become an essential tool for research and industrial applications. Given the large amount of high quality data needed and the often sensitive nature of ML data, privacy-preserving collaborative ML is of increasing importance. In this paper, we introduce new actively secure multiparty computation (MPC) protocols which are specially optimized for privacy-preserving machine learning applications. We concentrate on the optimization of (tensor) convolutions which belong to the most commonly used components in ML architectures, especially in convolutional neural networks but also in recurrent neural networks or transformers, and therefore have a major impact on the overall performance. Our approach is based on a generalized form of structured randomness that speeds up convolutions in a fast online phase. The structured randomness is generated with homomorphic encryption using adapted and newly constructed packing methods for convolutions, which might be of independent interest. Overall our protocols extend the state-of-the-art Overdrive family of protocols (Keller et al., EUROCRYPT 2018). We implemented our protocols on-top of MP-SPDZ (Keller, CCS 2020) resulting in a full-featured implementation with support for faster convolutions. Our evaluation shows that our protocols outperform state-of-the-art actively secure MPC protocols on ML tasks like evaluating ResNet50 by a factor of 3 or more. Benchmarks for depthwise convolutions show order-of-magnitude speed-ups compared to existing approaches.

The security of cryptographic primitives is an important issue. The Shor algorithm illustrates how quantum attacks threaten the security of these widely used primitives. Code-based cryptography is one of several approaches resistant to quantum attacks. To date, no attack has been able to break a code-based cryptosystem in polynomial time. Despite this level of security, these cryptosystems have not been considered for practical applications such as e-commerce, medical and industrial IoT, finance, blockchain, mobile services, and online banking. The main reason is the large public and private key sizes. This paper presents a new code-based cryptosystem based on inverse parity check matrices. The dual matrix provides both a parity check matrix transpose and a parity check matrix inverse. These are employed in the key generation, encryption, and decryption algorithms. The proposed scheme provides public and private key sizes smaller than the McEliece cryptosystem and has a higher level of security.

Fully homomorphic encryption FHE has been one of the most promising cryptographic tools for secure two-party computation and secure outsourcing computation in recent years. However, the complex bootstrapping procedure in FHE schemes is the main bottleneck of it practical usage, and the TFHE scheme is the state-of-the-art for efficient bootstrapping. To further improve the efficiency of bootstrapping in TFHE, the number of fast Fourier transforms (FFT) should be reduced since bootstrapping in TFHE is mainly composed of vast FFTs. In this paper, we focus on a novel method of decomposing-in-Fourier (DIF) to reduce the number of FFTs in bootstrapping of TFHE, from $2(\ell+1)n$ to $4n$. As a result, our method would reduce the number of FFTs required by each external product in bootstrapping to a constant number rather than varying with decomposing parameters, which leads to a scale-invariant bootstrapping structure.

IACR is soliciting for affiliated events to be held in conjunction with Crypto 2023 Saturday, August 19th, and/or Sunday, August 20th, 2023.

The call for proposals is now out at: https://crypto.iacr.org/2023/affiliated.php

Eurocrypt 2023 will take place in Lyon, France on April 23-27, 2023.

The registration site is now open. To avoid late fees please register before March 30th.
https://eurocrypt.iacr.org/2023/registration.php

The goal of this research is to raise technical doubts regarding the usefulness of the repeated attempts by governments to curb Cryptography (aka the ``Crypto Wars''), and argue that they, in fact, cause more damage than adding effective control. The notion of {\em Anamorphic Encryption} was presented in Eurocrypt'22 for a similar aim. There, despite the presence of a Dictator who possesses all keys and knows all messages, parties can arrange a hidden ``{\it anamorphic}'' message in an otherwise indistinguishable from regular ciphertexts (wrt the Dictator).

In this work, we postulate a stronger cryptographic control setting where encryption does not exist (or is neutralized) since all communication is passed through the Dictator in, essentially, cleartext mode (or otherwise, when secure channels to and from the Dictator are the only confidentiality mechanism). Messages are only authenticated to assure recipients of the identity of the sender. We ask whether security against the Dictator still exists, even under such a~strict regime which allows only authentication (i.e., authenticated/ signed messages) to pass end-to-end, and where received messages are determined by/ known to the Dictator, and the Dictator also eventually gets all keys to verify compliance of past signing. To frustrate the Dictator, this authenticated message setting gives rise to the possible notion of anamorphic channels inside signature and authentication schemes, where parties attempt to send undetectable secure messages (or other values) using signature tags which are indistinguishable from regular tags. We define and present implementation of schemes for anamorphic signature and authentication; these are applicable to existing and standardized signature and authentication schemes which were designed independently of the notion of anamorphic messages. Further, some cornerstone constructions of the foundations of signatures, in fact, introduce anamorphism.

MIBS is a 32-round lightweight block cipher following a Feistel structure with the block length of 64-bit and the key length of 64 or 80 bits. In this paper, the properties of the key scheduling algorithm are investigated and lots of repeated bits among the different round keys are found. Moreover, the optimal guessing order of the unknown key bits is obtained by using the greedy algorithm. At last, combined with the early abort technique, the differential cryptanalyses are improved to 15 rounds both of MIBS-64 and MIBS-80. For MIBS-64, the data complexity is $2^{59}$, and the time complexity is $2^{46.2}$ encryptions. For MIBS-80, the data complexity is $2^{59}$, and the time complexity is $2^{51.7}$ encryptions. The key scheduling algorithm of MIBS is similar to some other lightweight block ciphers, and we hope further similarities will help build better attacks for them as well.

GIFT-64 is a block cipher that has received a lot of attention from the community since its proposal in 2017. The attack on the highest number of rounds is a differential related-key attack on 26 rounds~\cite{DBLP:journals/tosc/SunWW21}. We studied this attack, in particular with respect to the generic framework for improving key recovery from~\cite{DBLP:conf/asiacrypt/BrollCFLN21}, and we realised that this framework, combined with an efficient parallel key guessing of interesting subsets of the key and a consequent list merging applied to the partial solutions, can improve the complexity of the attack. We propose two different trade-offs, as a result of the improved key-recovery. We believe that the techniques are quite generic and that it is possible to apply them to improve other differential attacks.

Over the years, a large number of attacks have been proposed against substitution boxes used in symmetric ciphers such as differential attacks, linear attacks, algebraic attacks, etc. In the Advanced Encryption Standard (AES) Block cipher, the substitution box is the only nonlinear component and thus it holds the weight of the cipher. This basically means that if an attacker is able to mount a successful attack on the substitution box of AES, the cipher is compromised. This research work aims to provide a solution for increasing cryptographic immunity of S-boxes against such attacks. A genetic algorithm based approach has been proposed to search for 8 × 8 balanced and bijective S-boxes that exhibit values of differential branch number, non-linearity, differential uniformity, count and length of cycles present and distance from strict avalanche criterion that are similar to or better than the AES S-box. An S-Box evaluation tool is also implemented to evaluate any S-boxes generated. S-box of AES is resistant to the crypt-analytic attacks. S-boxes constructed by the proposed algorithm have better cryptographic properties so they are also resistant to the crypt-analytic attacks. The strict avalanche criterion[11], which is based on completeness[22] and diffusion[5], is an essential property for any 8 × 8 S-box. Good diffusion means that a small change in the plaintext may influence the complete block after a small number of rounds. Therefore, a lower DSAC value is desirable to prevent vulnerabilities to attacks such as differential attacks. The DSAC is therefore used as the primary fitness criterion in this research work to search for S-boxes with better diffusion.

The Extended Access Control (EAC) protocol for authenticated key agreement is mainly used to secure connections between machine-readable travel documents (MRTDs) and inspection terminals, but it can also be adopted as a universal solution for attribute-based access control with smart cards. The security of EAC is currently based on the Diffie-Hellman problem, which may not be hard when considering quantum computers.

In this work we present PQ-EAC, a quantum-resistant version of the EAC protocol. We show how to achieve post-quantum confidentiality and authentication without sacrificing real-world usability on smart cards. To ease adoption, we present two main versions of PQ-EAC: One that uses signatures for authentication and one where authentication is facilitated using long-term KEM keys. Both versions can be adapted to achieve forward secrecy and to reduce round complexity. To ensure backwards-compatibility, PQ-EAC can be implemented using only Application Protocol Data Units (APDUs) specified for EAC in standard BSI TR-03110. Merely the protocol messages needed to achieve forward secrecy require an additional APDU not specified in TR-03110. We prove security of all versions in the real-or-random model of Bellare and Rogaway.

To show real-world practicality of PQ-EAC we have implemented a version using signatures on an ARM SC300 security controller, which is typically deployed in MRTDs. We also implemented PQ-EAC on a VISOCORE terminal for border control. We then conducted several experiments to evaluate the performance of PQ-EAC executed between chip and terminal under various real-world conditions. Our results strongly suggest that PQ-EAC is efficient enough for use in border control.

Herein, we propose an anonymous broadcast authentication (ABA) scheme to simultaneously control $10^9$ devices practically. We find a barrier to construct an ABA working with a larger number of devices.In a nutshell, there is a trilemma between (i) security, (ii) ciphertext length, and (iii) freedom in the target devices selection. For practical use, we propose ABAs with a ciphertext size of $O(\log N)$ where $N$ is the number of target devices while we impose a certain restriction on (iii). We provide an ABA template and instantiate it into specific schemes from the discrete logarithm problem (DLP) or the learning with errors (LWE) problem.