International Association
for Cryptologic Research

Transformer neural networks have gained significant traction since their introduction, becoming pivotal across diverse domains. Particularly in large language models like Claude and ChatGPT, the transformer architecture has demonstrated remarkable efficacy. This paper provides a concise overview of transformer neural networks and delves into their security considerations, focusing on covert channel attacks and their implications for the safety of large language models. We present a covert channel utilizing encryption and demonstrate its efficacy in circumventing Claude.ai's security measures. Our experiment reveals that Claude.ai appears to log our queries and blocks our attack within two days of our initial successful breach. This raises two concerns within the community: (1) The extensive logging of user inputs by large language models could pose privacy risks for users. (2) It may deter academic research on the security of such models due to the lack of experiment repeatability.

The Number Theoretic Transform (NTT) is a powerful mathematical tool that has become increasingly important in developing Post Quantum Cryptography (PQC) and Homomorphic Encryption (HE). Its ability to efficiently calculate polynomial multiplication using the convolution theorem with a quasi-linear complexity $O(n \log{n})$ instead of $O(n^2)$ when implemented with Fast Fourier Transform-style algorithms has made it a key component in modern cryptography. FFT-style NTT algorithm or fast-NTT is particularly useful in lattice-based cryptography. In this short note, we briefly introduce the basic concepts of linear, cyclic, and negacyclic convolutions via traditional schoolbook algorithms, traditional NTT, its inverse (INTT), and FFT-like versions of NTT/INTT. We then provide consistent toy examples through different concepts and algorithms to understand the basics of NTT concepts.

In the implementation of isogeny-based cryptographic schemes, Vélu’s formulas are essential for constructing and evaluating odd degree isogenies. Bernstein et al. proposed an approach known as √élu, which computes an ?-isogeny at a cost of̃ (√?) finite field operations. This paper presents two key improvements to enhance the efficiency of the implementation of √élu from two aspects: optimizing the partition involved in √élu and speeding up the computations of the sums of products used in polynomial multiplications over finite field ?? with large prime characteristic ?. To optimize the partition, we adjust it to enhance the utilization of ?-coordinates and eliminate the computational redundancy, which can ultimately reduce the number of ??-multiplications. The speedup of the sums of products is to employ two techniques: lazy reduction (abbreviated as LZYR) and generalized interleaved Montgomery multiplication (abbreviated as INTL). These techniques aim to minimize the underlying operations such as ??-reductions and assembly memory instructions. We present an optimized C and ssembly code implementation of √élu for the CTIDH512 instantiation. In terms of ?-isogeny computations in CTIDH512, the performance of clock cycles applying new partition + INTL (resp. new partition + LZYR) offers an improvement up to 16.05% (resp. 10.96%) compared to the previous work.

Recently, a paper by Chen (eprint 2024/555) has claimed to construct a quantum polynomial-time algorithm that solves the Learning With Errors Problem (Regev, JACM 2009), for a range of parameters. As a byproduct of Chen's result, it follows that Chen's algorithm solves the Gap Shortest Vector Problem, for gap $g(n) = \tilde{O}\left( n^{4.5} \right)$. In this short note we point to an error in the claims of Chen's paper.

We revisit the alternating moduli paradigm for constructing symmetric key primitives with a focus on constructing highly efficient protocols to evaluate them using secure multi-party computation (MPC). The alternating moduli paradigm of Boneh et al. (TCC 2018) enables the construction of various symmetric key primitives with the common characteristic that the inputs are multiplied by two linear maps over different moduli, first over $\mathbb{F}_2$ and then over $\mathbb{F}_3$.

The first contribution focuses on efficient two-party evaluation of alternating moduli PRFs, effectively building an oblivious pseudorandom function. We present a generalization of the PRF proposed by Boneh et al. (TCC 18) along with methods to lower the communication and computation. We then provide several variants of our protocols, with different computation and communication tradeoffs, for evaluating the PRF. Most are in the OT/VOLE hybrid model while one is based on specialized garbling. Our most efficient protocol effectively is about $3\times$ faster and requires $1.3\times$ lesser communication. %Moreover, our implementation is an order of magnitude faster than prior work. Our next contribution is the efficient evaluation of the OWF $f(x)=B\cdot_3 (A\cdot_2 x)$ proposed by Dinur et al. (CRYPTO 21) where $A \in \mathbb{F}^{m\times n}_2, B\in\mathbb{F}^{t\times m}_3$ and $\cdot_p$ is multiplication mod $p$. This surprisingly simple OWF can be evaluated within MPC by secret sharing $[\hspace{-3px}[x]\hspace{-3px}]$ over $\mathbb{F}_2$, locally computing $[\hspace{-3px}[v]\hspace{-3px}]=A\cdot_2 [\hspace{-3px}[x]\hspace{-3px}]$, performing a modulus switching protocol to $\mathbb{F}_3$ shares, followed by locally computing the output shares $[\hspace{-3px}[y]\hspace{-3px}]=B\cdot_3 [\hspace{-3px}[v]\hspace{-3px}]$. %While Dinur et al. used the KKW framework (CCS 2018) to construct a post quantum signature, w We design a bespoke MPC-in-the-Head (MPCitH) signature scheme that evaluates the OWF, achieving state of art performance. The resulting signature has a size ranging from 4.0-5.5 KB, achieving between $2\text{-}3\times$ reduction compared to Dinur et al. To the best of our knowledge, this is only $\approx 5\%$ larger than the smallest signature based on symmetric key primitives, including the latest NIST PQC competition submissions. We additionally show that our core techniques can be extended to build very small post-quantum ring signatures for small-medium sized rings that are competitive with state-of-the-art lattice based schemes. Our techniques are in fact more generally applicable to set membership in MPCitH.

In this paper, we introduce the first fault attack on SQIsign. By injecting a fault into the ideal generator during the commitment phase, we demonstrate a meaningful probability of inducing the generation of order $\mathcal{O}_0$. The probability is bounded by one parameter, the degree of commitment isogeny. We also show that the probability can be reasonably estimated by assuming uniform randomness of a random variable, and provide empirical evidence supporting the validity of this approximation. In addition, we identify a loop-abort vulnerability due to the iterative structure of the isogeny operation. Exploiting these vulnerabilities, we present key recovery fault attack scenarios for two versions of SQIsign---one deterministic and the other randomized. We then analyze the time complexity and the number of queries required for each attack. Finally, we discuss straightforward countermeasures that can be implemented against the attack.

Dynamic Decentralized Functional Encryption (DDFE), introduced by Chotard et al. (CRYPTO'20), stands as a robust generalization of (Multi-Client) Functional Encryption. It enables users to dynamically join and contribute private inputs to individually-controlled joint functions, all without requiring a trusted authority. Agrawal et al. (TCC’21) further extended this line of research by presenting the first DDFE construction for function-hiding inner products (FH-IP-DDFE) in the random oracle model (ROM).

Recently, Shi et al. (PKC'23) proposed the first Multi-Client Functional Encryption construction for function-hiding inner products based on standard assumptions without using random oracles. However, their construction still necessitates a trusted authority, leaving the question of whether a fully-fledged FH-IP-DDFE can exist in the standard model as an exciting open problem.

In this work, we provide an affirmative answer to this question by proposing a FH-IP-DDFE construction based on the Symmetric External Diffie-Hellman (SXDH) assumption in the standard model. Our approach relies on a novel zero-sharing scheme termed Updatable Pseudorandom Zero Sharing, which introduces new properties related to updatability in both definition and security models. We further instantiate this scheme in groups where the Decisional Diffie-Hellman (DDH) assumption holds.

Moreover, our proposed pseudorandom zero sharing scheme serves as a versatile tool to enhance the security of pairing-based DDFE constructions for functionalities beyond inner products. As a concrete example, we present the first DDFE for attribute-weighted sums in the standard model, complementing the recent ROM-based construction by Agrawal et al. (CRYPTO'23).

The Ascon cipher suite has recently become the preferred standard in the NIST Lightweight Cryptography standardization process. Despite its prominence, the initial dedicated security analysis for the Ascon mode was conducted quite recently. This analysis demonstrated that the Ascon AEAD mode offers superior security compared to the generic Duplex mode, but it was limited to a specific scenario: single-user nonce-respecting, with a capacity strictly larger than the key size. In this paper, we eliminate these constraints and provide a comprehensive security analysis of the Ascon AEAD mode in the multi-user setting, where the capacity need not be larger than the key size. Regarding data complexity $D$ and time complexity $T$, our analysis reveals that Ascon achieves AEAD security when $T$ is bounded by $\min\{2^{\kappa}/\mu, 2^c\}$ (where $\kappa$ is the key size, and $\mu$ is the number of users), and $DT$ is limited to $2^b$ (with $b$ denoting the size of the underlying permutation, set at 320 for Ascon). Our results align with NIST requirements, showing that Ascon allows for a tag size as small as 64 bits while supporting a higher rate of 192 bits, provided the number of users remains within recommended limits. However, this security becomes compromised as the number of users increases significantly. To address this issue, we propose a variant of the Ascon mode called LK-Ascon, which enables doubling the key size. This adjustment allows for a greater number of users without sacrificing security, while possibly offering additional resilience against quantum key recovery attacks. We establish tight bounds for LK-Ascon, and furthermore show that both Ascon and LK-Ascon maintain authenticity security even when facing nonce-misuse adversaries.

In this paper we address the use of Neural Networks (NN) for the assessment of the quality and hence safety of several Random Number Generators (RNGs), focusing both on the vulnerability of classical Pseudo Random Number Generators (PRNGs), such as Linear Congruential Generators (LCGs) and the RC4 algorithm, and extending our analysis to non-conventional data sources, such as Quantum Random Number Generators (QRNGs) based on Vertical-Cavity Surface- Emitting Laser (VCSEL). Among the results found, we identified a sort of classification of generators under different degrees of susceptibility, underlining the fundamental role of design decisions in enhancing the safety of PRNGs. The influence of network architecture design and associated hyper-parameters variations was also explored, highlighting the effectiveness of longer sequence lengths and convolutional neural networks in enhancing the discrimination of PRNGs against other RNGs. Moreover, in the prediction domain, the proposed model is able to deftly distinguish the raw data of our QRNG from truly random ones, exhibiting a cross-entropy error of 0.52 on the test data-set used. All these findings reveal the potential of NNs to enhance the security of RNGs, while highlighting the robustness of certain QRNGs, in particular the VCSEL-based variants, for high-quality random number generation applications.

In symmetric cryptography, vectorial Boolean functions over finite fields F2n derive strong S-boxes. To this end, the S-box should satisfy a list of tests to resist existing attacks, such as the differential, linear, boomerang, and variants. Several tables are employed to measure an S- box’s resistance, such as the difference distribution table (DDT) and the boomerang connectivity table (BCT). Following the boomerang attacks recently revisited in terms of the boomerang switch effect, with a lustra- tion highlighting the power of this technique, a tool called the Boomerang Difference Table (BDT), an alternative to the classical Boomerang BCT, was introduced. Next, two novel tables have been introduced, namely, the Upper Boomerang Connectivity Table (UBCT) and the Lower Boomerang Connectivity Table (LBCT), which are considered improvements over BCT while allowing systematic evaluation of boomerangs to return over mul- tiple rounds. This paper focuses on the new tools for measuring the revisited version of boomerang attacks and the related tables UBCT, LBCT, as well as the so-called Extended Boomerang Connectivity Table (EBCT). Specifically, we shall study the properties of these novel tools and investigate the corresponding tables. We also study their interconnections, their links to the DDT, and their values for affine equivalent vectorial functions and compositional inverses of permutations of F2n . Moreover, we introduce the concept of the nontrivial boomerang connectivity uniformity and determine the explicit values of all the entries of the EBCT, LBCT, and EBCT for the important cryptographic case of the inverse function.

In this paper, we investigate the computational complexity of the problem of solving a one-sided system of equations of degree two of a special form over the max-plus algebra. Also, we consider the asymptotic density of solvable systems of this form. Such systems have appeared during the analysis of some tropical cryptography protocols that were recently suggested. We show how this problem is related to the integer linear programming problem and prove that this problem is NP-complete. We show that the asymptotic density of solvable systems of this form with some restrictions on the coefficients, the number of variables, and the number of equations is 0. As a corollary, we prove that this problem (with some restrictions on the coefficients, the number of variables, and the number of equations) is decidable generically in polynomial time.

In isogeny-based cryptography, bilinear pairings are regarded as a powerful tool in various applications, including key compression, public-key validation and torsion basis generation. However, in most isogeny-based protocols, the performance of pairing computations is unsatisfactory due to the high computational cost of the Miller function. Reducing the computational expense of the Miller function is crucial for enhancing the overall performance of pairing computations in isogeny-based cryptography.

This paper addresses this efficiency bottleneck. To achieve this, we propose several techniques for a better implementation of pairings in isogeny-based cryptosystems. We use (modified) Jacobian coordinates and present new algorithms for Miller function computations to compute pairings of order $2^\bullet$ and $3^\bullet$. For pairings of arbitrary order, which are crucial for key compression in some SIDH-based schemes (such as M-SIDH and binSIDH), we combine Miller doublings with Miller additions/subtractions, leading to a considerable speedup. Moreover, the optimizations for pairing applications in CSIDH-based protocols are also considered in this paper. In particular, our approach for supersingularity verification in CSIDH is 15.3% faster than Doliskani's test, which is the state-of-the-art.

Software solutions to address computational challenges are ubiquitous in our daily lives. One specific application area where software is often used is in embedded systems, which, like other digital electronic devices, are vulnerable to side-channel analysis attacks. Although masking is the most common countermeasure and provides a solid theoretical foundation for ensuring security, recent research has revealed a crucial gap between theoretical and real-world security. This shortcoming stems from the micro-architectural effects of the underlying micro-processor. Common security models used to formally verify masking schemes such as the d-probing model fully ignore the micro-architectural leakages that lead to a set of instructions that unintentionally recombine the shares. Manual generation of masked assembly code that remains secure in the presence of such micro-architectural recombinations often involves trial and error, and is non-trivial even for experts. Motivated by this, we present PoMMES, which enables inexperienced software developers to automatically compile masked functions written in a high-level programming language into assembly code, while preserving the theoretically proven security in practice. Compared to the state of the art, based on a general model for microarchitectural effects, our scheme allows the generation of practically secure masked software at arbitrary security orders for in-order processors. The major contribution of PoMMES is its micro-architecture aware register allocation algorithm, which is one of the crucial steps during the compilation process. In addition to simulation-based assessments that we conducted by open-source tools dedicated to evaluating masked software implementations, we confirm the effectiveness of the PoMMES-generated codes through experimental analysis. We present the result of power consumption based leakage assessments of several case studies running on a Cortex M0+ micro-controller, which is commonly deployed in industry.

Searchable Symmetric Encryption (SSE) has opened up an attractive avenue for privacy-preserved processing of outsourced data on the untrusted cloud infrastructure. SSE aims to support efficient Boolean query processing with optimal storage and search overhead over large real databases. However, current constructions in the literature lack the support for multi-client search and dynamic updates to the encrypted databases, which are essential requirements for the widespread deployment of SSE on real cloud infrastructures. Trivially extending a state-of-the-art single client dynamic construction, such as ODXT (Patranabis et al., NDSS’21), incurs significant leakage that renders such extension insecure in practice. Currently, no SSE construction in the literature offers efficient multi-client query processing and search with dynamic updates over large real databases while maintaining a benign leakage profile. This work presents the first dynamic multi-client SSE scheme Nomos supporting efficient multi-client conjunctive Boolean queries over an encrypted database. Precisely, Nomos is a multi-reader-single-writer construction that allows only the gate-keeper (or the data-owner) - a trusted entity in the Nomos framework, to update the encrypted database stored on the adversarial server. Nomos achieves forward and type-II backward privacy of dynamic SSE constructions while incurring lesser leakage than the trivial extension of ODXT to a multi- client setting. Furthermore, our construction is practically efficient and scalable - attaining linear encrypted storage and sublinear search overhead for conjunctive Boolean queries. We provide an experimental evaluation of software implementation over an extensive real dataset containing millions of records. The results show that Nomos performance is comparable to the state-of-the-art static conjunctive SSE schemes in practice.

Satisfiability modulo finite fields enables automated verification for cryptosystems. Unfortunately, previous solvers scale poorly for even some simple systems of field equations, in part because they build a full Gröbner basis (GB) for the system. We propose a new solver that uses multiple, simpler GBs instead of one full GB. Our solver, implemented within the cvc5 SMT solver, admits specialized propagation algorithms, e.g., for understanding bitsums. Experiments show that it solves important bitsum-heavy determinism benchmarks far faster than prior solvers, without introducing much overhead for other benchmarks.

We give a new protocol for reliable broadcast with improved communication complexity for long messages. Namely, to reliably broadcast a message a message $m$ over an asynchronous network to a set of $n$ parties, of which fewer than $n/3$ may be corrupt, our protocol achieves a communication complexity of $1.5 |m| n + O( \kappa n^2 \log(n) )$, where $\kappa$ is the output length of a collision-resistant hash function. This result improves on the previously best known bound for long messages of $2 |m| n + O( \kappa n^2 \log(n) )$.

Private set intersection (PSI) allows two parties to compute the intersection of their sets without revealing anything else. In some applications of PSI, a server holds a large set and needs to run PSI with many clients, each with its own small set. In this setting, however, all existing protocols fall short: they either incur too much cost to compute the intersections for many clients or cannot achieve the desired security requirements.

We design a protocol that particularly suits this setting with simulation-based security against malicious adversaries. In our protocol, the server publishes a one-time, linear-size encoding of its set. Then, multiple clients can each perform a cheap interaction with the server of complexity linear in the size of each client's set. A key ingredient of our protocol is an efficient instantiation of an oblivious verifiable unpredictable function, which could be of independent interest. To obtain the intersection, the client can download the encodings directly, which can be accelerated via content distribution networks or peer-to-peer networks since the same encoding is used by all clients; alternatively, clients can fetch only the relevant ones using verifiable private information retrieval.

Our implementation shows very high efficiency. For a server holding $10^8$ elements and each client holding $10^3$ elements, the size of the server's encoding is 800MB; interacting with each client uses 60MB of communication and runs in under 5s in a WAN network with 120Mbps bandwidth. Compared with the state-of-the-art PSI protocol, our scheme requires only 0.017 USD per client on an AWS server, which is 5x lower.

The IACR Fellows Program recognizes outstanding IACR members for technical and professional contributions to the field of cryptology. Today we are pleased to announce eight members that have been elevated to the rank of Fellow for 2024:

• Anne Canteaut, for influential contributions to symmetric cryptography and Boolean functions, and for exemplary service to the symmetric cryptography community.
• Joan Feigenbaum, for highly influential contributions to the foundations of trust and secure computation, and for service to the IACR.
• Alfred Menezes, for fundamental contributions to the theory and practice of elliptic curve cryptography, and for service to the cryptographic community.
• Kobbi Nissim, for fundamental contributions to the theory and practice of data privacy, and for service to the cryptographic community.
• Chris Peikert, for fundamental contributions to the functionality, efficiency, and security of lattice-based cryptography, and for service to the IACR.
• David Pointcheval, for fundamental contributions to the design of public-key cryptosystems and their provable security analysis, for educational leadership, and for outstanding service to the IACR.
• François-Xavier Standaert, for fundamental contributions to the theory and practice of cryptography in the presence of leakage, and for service to the IACR.
• Brent Waters, for the development of attribute-based encryption, functional encryption, and other foundational concepts in cryptography, and for service to the cryptographic community.

Congratulations to the new fellows! More information about the IACR Fellows Program can be found at https://iacr.org/fellows/.

This postdoctoral position is for 3 years (with possible extension to up to 4 years) and is linked with a research effort aiming to apply cryptographic methods for studying the security aspects of Artificial Intelligence (AI). Cryptographic approaches will be tested for attacks on AI and existing protection methods in cryptography will be applied for security of AI. The candidate will have the freedom to explore new related topics and drive the research in the direction of security of AI and cryptographic elements. The postdoc will work with a team at the Selmer Center in Secure Communication, including among others Adj. Prof. Vincent Rijmen, Adj. Prof. Sondre Rønjom and Prof. Lilya Budaghyan. Please apply online via the link where more information is available: https://www.jobbnorge.no/en/available-jobs/job/260444/lead-ai-postdoctoral-research-fellow-position-within-cryptography-and-security-of-ai

Closing date for applications:

Contact: Prof. Budaghyan

More information: https://www.jobbnorge.no/en/available-jobs/job/260444/lead-ai-postdoctoral-research-fellow-position-within-cryptography-and-security-of-ai

Fuzzy password authenticated key exchange (fuzzy PAKE) protocols enable two parties to securely exchange a session-key for further communication. The parties only need to share a low entropy password. The passwords do not even need to be identical, but can contain some errors. This may be due to typos, or because the passwords were created from noisy biometric readings. In this paper we provide an overview and comparison of existing fuzzy PAKE protocols. Furthermore, we analyze certain security properties of these protocols and argue that the protocols can be expected to be slightly more secure in practice than can be inferred from their theoretical guarantees.