International Association
for Cryptologic Research

We propose an efficient non-interactive privacy-preserving Transformer inference architecture called Powerformer. Since softmax is a non-algebraic operation, previous studies have attempted to modify it to be HE-friendly, but these methods have encountered issues with accuracy degradation or prolonged execution times due to the use of multiple bootstrappings. We propose replacing softmax with a new ReLU-based function called the \textit{Batch Rectifier-Power max} (BRPmax) function without any unstable approximation methods, which outperforms even original BERT performance within BERT-Large model while requiring fewer levels, allowing it to operate with only a single bootstrapping. We also present a matrix multiplication algorithms specialized for attention block that reduce the number of key-switchings by 35% to 91% compared to existing state-of-the-art methods. We design clear end-to-end HE-based implementation for private Transformer model, and our implementation of Powerformer on the BERT-tiny model using RNS-CKKS takes 503 seconds on a single-threaded CPU, and to the best of our knowledge, this is the first end-to-end non-interactive Transformer implementation using HE.

Federated Learning (FL) enables multiple clients to collaboratively train a machine learning model while keeping their data private, eliminating the need for data sharing. Two common approaches to secure aggregation (SA) in FL are the single-aggregator and multiple-aggregator models. Existing multiple-aggregator protocols such as Prio (NSDI 2017), Prio+ (SCN 2022), Elsa (S\&P 2023) either offer robustness only in the presence of semi-honest servers or provide security without robustness and are limited to two aggregators. We introduce Mario, the first multi-aggregator SA protocol that is both secure in a malicious setting and provides robustness. Similar to prior work of Prio and Prio+, Mario provides secure aggregation in a setup of $n$ servers and $m$ clients. Unlike previous work, Mario removes the assumption of semi-honest servers, and provides a complete protocol with robustness against less than $n/2$ malicious servers, defense with input validation of upto $m-2$ corrupted clients, and dropout of any number of clients. Our implementation shows that Mario is $3.40\times$ and $283.4\times$ faster than Elsa and Prio+, respecitively.

In the Zero-Knowledge Proof (ZKP) of a disjunctive statement, $\mathcal{P}$ and $\mathcal{V}$ agree on $B$ fan-in $2$ circuits $\mathcal{C}_0, \ldots, \mathcal{C}_{B-1}$ over a field $\mathbb{F}$; each circuit has $n_{\mathit{in}}$ inputs, $n_\times$ multiplications, and one output. $\mathcal{P}$'s goal is to demonstrate the knowledge of a witness $(\mathit{id} \in [B]$, $\boldsymbol{w} \in \mathbb{F}^{n_{\mathit{in}}})$, s.t. $\mathcal{C}_{\mathit{id}}(\boldsymbol{w}) = 0$ where neither $\boldsymbol{w}$ nor $\mathit{id}$ is revealed. Disjunctive statements are effective, for example, in implementing ZKP based on sequential execution of CPU steps.

This paper studies ZKP (of knowledge) protocols over disjunctive statements based on Vector OLE. Denoting by $\lambda$ the statistical security parameter and let $\rho \overset{\Delta}{=} \max\{\log |\mathbb{F}|, \lambda\}$, the previous state-of-the-art protocol $\mathsf{Robin}$ (Yang et al. CCS'23) required $(n_{\mathit{in}}+3n_\times)\log \left|\mathbb{F}\right| + \mathcal{O}(\rho B)$ bits of communication with $ \mathcal{O}(1)$ rounds, and $\mathsf{Mac'n'Cheese}$ (Baum et al. CRYPTO'21) required $(n_{\mathit{in}}+n_\times)\log \left|\mathbb{F}\right| + 2n_\times\rho + \mathcal{O}(\rho \log B)$ bits of communication with $\mathcal{O}(\log B)$ rounds, both in the VOLE-hybrid model.

Our novel protocol $\mathsf{LogRobin}\texttt{++}$ achieves the same functionality at the cost of $(n_{\mathit{in}}+n_\times)\log \left|\mathbb{F}\right| + \mathcal{O}(\rho \log B)$ bits of communication with $\mathcal{O}(1)$ rounds in the VOLE-hybrid model. Crucially, $\mathsf{LogRobin}\texttt{++}$ takes advantage of two new techniques -- (1) an $\mathcal{O}(\log B)$-overhead approach to prove in ZK that an IT-MAC commitment vector contains a zero; and (2) the realization of VOLE-based ZK over a disjunctive statement, where $\mathcal{P}$ commits only to $\boldsymbol{w}$ and multiplication outputs of $\mathcal{C}_{\mathit{id}}(\boldsymbol{w})$ (as opposed to prior work where $\mathcal{P}$ commits to $\boldsymbol{w}$ and all three wires that are associated with each multiplication gate).

We implemented $\mathsf{LogRobin}\texttt{++}$ over Boolean (i.e., $\mathbb{F}_2$) and arithmetic (i.e., $\mathbb{F}_{2^{61}-1}$) fields. In our experiments, including the cost of generating VOLE correlations, $\mathsf{LogRobin}\texttt{++}$ achieved up to $170 \times$ optimization over $\mathsf{Robin}$ in communication, resulting in up to $7\times$ (resp. $3\times$) wall-clock time improvements in a WAN-like (resp. LAN-like) setting.

Cryptographic agility, the ability to easily and quickly modify cryptography in a sys- tem, is one of the most important features of any cryptographic system. Any algorithm may be attacked and, at some point in time, be broken. The most obvious solution is to change the cryptographic algorithm, however this has high risk and cost. Another solution is to use agile algorithms. Agile algorithms have security parameters easily changed to increase protection against attacks. In this paper we will show that finite field based algorithms are the most agile of currently used classical cryptography. A critical portion of this will be to show that the bottleneck for the primary costing attack, the number field sieve, is the linear algebra portion of the attack, and not the sieving portion. This paper examines the agility of all three algorithm categories and dispels a few myths about their strengths.

Introduced in [CG24], pseudorandom error-correcting codes (PRCs) are a new cryptographic primitive with applications in watermarking generative AI models. These are codes where a collection of polynomially many codewords is computationally indistinguishable from random, except to individuals with the decoding key. In this work, we examine the assumptions under which PRCs with robustness to a constant error rate exist. 1. We show that if both the planted hyperloop assumption introduced in [BKR23] and security of a version of Goldreich's PRG hold, then there exist public-key PRCs for which no efficient adversary can distinguish a polynomial number of codewords from random with better than $o(1)$ advantage. 2. We revisit the construction of [CG24] and show that it can be based on a wider range of assumptions than presented in [CG24]. To do this, we introduce a weakened version of the planted XOR assumption which we call the weak planted XOR assumption and which may be of independent interest. 3. We initiate the study of PRCs which are secure against space-bounded adversaries. We show how to construct secret-key PRCs of length $O(n)$ which are $\textit{unconditionally}$ indistinguishable from random by $\text{poly}(n)$ time, $O(n^{1.5-\varepsilon})$ space adversaries.

We propose waterlogs, a new direction to detect and trace synthetic text outputs from large language models based on transparency logs. Waterlogs offer major categorical advantages over watermarking: it (1) allows for the inclusion of arbitrary metadata to facilitate tracing, (2) is publicly verifiable by third parties, and (3) operates in a distributed manner while remaining robust and efficient.

Waterlogs rely on a verifiable Hamming distance index, a novel data structure that we describe, to efficiently search multi-dimensional semantic hashes of natural language embeddings in a verifiable manner. This data structure may be of independent interest.

We implement a waterlog, which we call DREDGE, and benchmark it using synthetic text generated by GPT-2 1.5B and OPT-13B; embeddings are generated via OpenAI's text-embedding-ada-002 model. We provide empirical benchmarks on the efficiency of appending text to the log and querying it for matches. We compare our results to watermarking and outline areas for further research.

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.

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.

This paper studies the provable security of the deterministic random bit generator~(DRBG) utilized in Linux 6.4.8, marking the first analysis of Linux-DRBG from a provable security perspective since its substantial structural changes in Linux 4 and Linux 5.17. Specifically, we prove its security up to $O(\min\{2^{\frac{n}{2}},2^{\frac{\lambda}{2}}\})$ queries in the seedless robustness model, where $n$ is the output size of the internal primitives and $\lambda$ is the min-entropy of the entropy source. Our result implies $128$-bit security given $n=256$ and $\lambda=256$ for Linux-DRBG. We also present two distinguishing attacks using $O(2^{\frac{n}{2}})$ and $O (2^{\frac{\lambda}{2}})$ queries, respectively, proving the tightness of our security bound.

We present novel Secure Multi-Party Computation (SMPC) protocols to perform Breadth-First-Searches (BFSs) and determine maximal flows on dense secret-shared graphs. In particular, we introduce a novel BFS protocol that requires only $\mathcal{O}(\log n)$ communication rounds on graphs with $n$ nodes, which is a big step from prior work that requires $\mathcal{O}(n \log n)$ rounds. This BFS protocol is then used in a maximal flow protocol based on the Edmonds-Karp algorithm, which requires $\mathcal{O}(n^3 \log n)$ rounds. We further optimize the protocol for cases where an upper bound $U$ on the capacities is publicly known by using a capacity scaling approach. This yields a new protocol which requires $\mathcal{O}(n^2 \log n \log U)$ rounds. Finally, we introduce a novel max flow protocol based on algorithms by Dinic and Tarjan with round complexity $\mathcal{O}(n^3)$.

All protocols presented in this paper use SMPC primitives as a black-box, allowing our protocols to be used as building blocks in a wide range of settings and applications. We evaluate our protocols with semi-honest and malicious security in different network settings. Our logarithmic BFS protocol is up to 69 times faster than prior protocols on small graphs with less than 100 nodes, but is outperformed by protocols with lower computational complexity on graphs with thousands of nodes. Further, we find our Dinic-Tarjan protocol to be faster than the Edmonds-Karp and capacity scaling protocols in our evaluation, albeit trends indicating capacity scaling protocols to be faster on graph sizes not reached in our evaluation.

Recently, Masny and Rindal [MR19] formalized a notion called Endemic Oblivious Transfer (EOT), and they proposed a generic transformation from Non-Interactive Key Exchange (NIKE) to EOT with standalone security in the random oracle (RO) model. However, from the model level, the relationship between idealized NIKE and idealized EOT and the relationship between idealized elementary public key primitives have been rarely researched. In this work, we investigate the relationship between ideal NIKE and ideal one-round EOT, as well as the relationship between ideal public key encryption (PKE) and ideal two-round Oblivious Transfer (OT), in the indifferentiability framework proposed by Maurer et al.(MRH04). Our results are threefold: Firstly, we model ideal PKE without public key validity test, ideal one-round EOT and ideal two-round OT in the indifferentiability framework. Secondly, we show that ideal NIKE and ideal one-round EOT are equivalent, and ideal PKE without public key validity test are equivalent to ideal two-round OT. Thirdly, we show a separation between ideal two-round OT and ideal one-round EOT, which implies a separation between ideal PKE and ideal NIKE.

We obfuscate words of a given length in a free monoid on two generators with a simple factorization algorithm (namely $SL_2(\mathbb{N})$) to create a public-key encryption scheme. We provide a reference implementation in Python and suggested parameters. The security analysis is between weak and non-existent, left to future work.

A broadcast encryption scheme allows a user to encrypt a message to $N$ recipients with a ciphertext whose size scales sublinearly with $N$. While broadcast encryption enables succinct encrypted broadcasts, it also introduces a strong trust assumption and a single point of failure; namely, there is a central authority who generates the decryption keys for all users in the system. Distributed broadcast encryption offers an appealing alternative where there is a one-time (trusted) setup process that generates a set of public parameters. Thereafter, users can independently generate their own public keys and post them to a public-key directory. Moreover, anyone can broadcast an encrypted message to any subset of user public keys with a ciphertext whose size scales sublinearly with the size of the broadcast set. Unlike traditional broadcast encryption, there are no long-term secrets in distributed broadcast encryption and users can join the system at any time (by posting their public key to the public-key directory).

Previously, distributed broadcast encryption schemes were known from standard pairing-based assumptions or from powerful tools like indistinguishability obfuscation or witness encryption. In this work, we provide the first distributed broadcast encryption scheme from a falsifiable lattice assumption. Specifically, we rely on the $\ell$-succinct learning with errors (LWE) assumption introduced by Wee (CRYPTO 2024). Previously, the only lattice-based candidate for distributed broadcast encryption goes through general-purpose witness encryption, which in turn is only known from the /private-coin/ evasive LWE assumption, a strong and non-falsifiable lattice assumption. Along the way, we also describe a more direct construction of broadcast encryption from lattices.

We present new lattice-based attribute-based encryption (ABE) and laconic function evaluation (LFE) schemes for circuits with *sublinear* ciphertext overhead. For depth $d$ circuits over $\ell$-bit inputs, we obtain

* an ABE with ciphertext and secret key size $O(1)$;

* a LFE with ciphertext size $\ell + O(1)$ and digest size $O(1)$;

* an ABE with public key and ciphertext size $O(\ell^{2/3})$ and secret key size $O(1)$,

where $O(\cdot)$ hides $\mbox{poly}(d,\lambda)$ factors. The first two results achieve almost optimal ciphertext and secret key / digest sizes, up to the $\mbox{poly}(d)$ dependencies. The security of our schemes relies on $\ell$-succinct LWE, a falsifiable assumption which is implied by evasive LWE. At the core of our results is a new technique for compressing LWE samples $\mathbf{s}(\mathbf{A}-\mathbf{x} \otimes \mathbf{G})$ as well as the matrix $\mathbf{A}$.

Light clients implement a simple solution for Bitcoin's scalability problem, as they do not store the entire blockchain but only the state of particular addresses of interest. To be able to keep track of the updated state of their addresses, light clients rely on full nodes to provide them with the required information. To do so, they must reveal information about the addresses they are interested in. This paper studies the two most common light client implementations, SPV and Neutrino with regards to their privacy. We define privacy metrics for comparing the privacy of the different implementations. We evaluate and compare the privacy of the implementations over time on real Bitcoin data and discuss the inherent privacy-communication tradeoff. In addition, we propose general techniques to enhance light client privacy in the existing implementations. Finally, we propose a new SPV-based light client model, the aggregation model, evaluate it, and show it can achieve enhanced privacy than in the existing light client implementations.

Zero-Knowledge (ZK) protocols allow a prover to demonstrate the truth of a statement without disclosing additional information about the underlying witness. Code-based cryptography has a long history but did suffer from periods of slow development. Recently, a prominent line of research have been contributing to designing efficient code-based ZK from MPC-in-the-head (Ishai et al., STOC 2007) and VOLE-in-the head (VOLEitH) (Baum et al., Crypto 2023) paradigms, resulting in quite efficient standard signatures. However, none of them could be directly used to construct privacy-preserving cryptographic primitives. Therefore, Stern's protocols remain to be the major technical stepping stones for developing advanced code-based privacy-preserving systems.

This work proposes new code-based ZK protocols from VOLEitH paradigm for various relations and designs several code-based privacy-preserving systems that considerably advance the state-of-the-art in code-based cryptography. Our first contribution is a new ZK protocol for proving the correctness of a regular (non-linear) encoding process, which is utilized in many advanced privacy-preserving systems. Our second contribution are new ZK protocols for concrete code-based relations. In particular, we provide a ZK of accumulated values with optimal witness size for the accumulator (Nguyen et al., Asiacrypt 2019). Our protocols thus open the door for constructing more efficient privacy-preserving systems. Moreover, our ZK protocols have the advantage of being simpler, faster, and smaller compared to Stern-like protocols. To illustrate the effectiveness of our new ZK protocols, we develop ring signature (RS) scheme, group signature (GS) scheme, fully dynamic attribute-based signature scheme from our new ZK. The signature sizes of the resulting schemes are two to three orders of magnitude smaller than those based on Stern-like protocols in various parameter settings. Finally, our first ZK protocol yields a standard signature scheme, achieving ``signature size + public key size'' as small as $3.05$ KB, which is slightly smaller than the state-of-the-art signature scheme (Cui et al., PKC 2024) based on the regular syndrome decoding problems.

Zero-Knowledge (ZK) protocols have been a subject of intensive study due to their fundamental importance and versatility in modern cryptography. However, the inherently different nature of quantum information significantly alters the landscape, necessitating a re-examination of ZK designs.

A crucial aspect of ZK protocols is their round complexity, intricately linked to $\textit{simulation}$, which forms the foundation of their formal definition and security proofs. In the $\textit{post-quantum}$ setting, where honest parties and their communication channels are all classical but the adversaries could be quantum, Chia, Chung, Liu, and Yamakawa [FOCS'21] demonstrated the non-existence of constant-round $\textit{black-box-simulatable}$ ZK arguments (BBZK) for $\mathbf{NP}$ unless $\mathbf{NP} \subseteq \mathbf{BQP}$. However, this problem remains widely open in the full-fledged quantum future that will eventually arrive, where all parties (including the honest ones) and their communication are naturally quantum.

Indeed, this problem is of interest to the broader theory of quantum computing. It has been an important theme to investigate how quantum power fundamentally alters traditional computational tasks, such as the $\textit{unconditional}$ security of Quantum Key Distribution and the incorporation of Oblivious Transfers in MiniQCrypt. Moreover, quantum communication has led to round compression for commitments and interactive arguments. Along this line, the above problem is of great significance in understanding whether quantum computing could also change the nature of ZK protocols in some fundamentally manner.

We resolved this problem by proving that only languages in $\mathbf{BQP}$ admit constant-round $\textit{fully-quantum}$ BBZK. This result holds significant implications. Firstly, it illuminates the nature of quantum zero-knowledge and provides valuable insights for designing future protocols in the quantum realm. Secondly, it relates ZK round complexity with the intriguing problem of $\mathbf{BQP}$ vs $\mathbf{QMA}$, which is out of the reach of previous analogue impossibility results in the classical or post-quantum setting. Lastly, it justifies the need for the $\textit{non-black-box}$ simulation techniques or the relaxed security notions employed in existing constant-round fully-quantum BBZK protocols.

Let $\zeta(z)=\sum_{n=1}^{\infty} \frac{1}{n^z}$, $\psi(z)=\sum_{n=1}^{\infty} \frac{(-1)^{n-1}}{n^z}, z\in \mathbb{C}$. We show that $\psi(z)\not=(1-2^{1-z})\zeta(z)$, if $0

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.

This work introduces Cryptobazaar, a novel scalable, private, and decentralized sealed-bid auction protocol. In particular, our protocol protects the privacy of losing bidders by preserving the confidentiality of their bids while ensuring public verifiability of the outcome and relying only on a single untrusted auctioneer for coordination. At its core, Cryptobazaar combines an efficient distributed protocol to compute the logical-OR for a list of unary-encoded bids with various novel zero-knowledge succinct arguments of knowledge that may be of independent interest. We further present variants of our protocol that can be used for efficient first-, second-, and more generally $(p+1)$st-price as well as sequential first-price auctions. Finally, the performance evaluation of our Cryptobazaar implementation shows that it is highly practical. For example, a single run of an auction with $128$ bidders and a price range of $1024$ values terminates in less than $0.5$ sec and requires each bidder to send and receive only about $32$ KB of data.