International Association
for Cryptologic Research

Federated fine-tuning of Large Language Models via Low Rank Adaptation (LoRA) faces a critical privacy-efficiency trade-off: low-rank factors can leak sensitive data, yet standard secure aggregation is restricted to linear operations. Existing solutions for aggregating matrix products (e.g., $\mathbf{B}_i \mathbf{A}_i$) either sacrifice exactness, depend on a trusted third party, or incur prohibitive costs at scale. We present SecLoRA, the first decentralized framework achieving exact aggregation of LoRA updates with linear communication complexity. The core of SecLoRA is a novel cryptographic primitive: Pairwise Composable Multi-Client Functional Encryption (PC-DMCFE). Unlike traditional functional encryption, which treats ciphertext recombination as an attack, the dual-encryption architecture of PC-DMCFE ($\mathsf{Enc}_A, \mathsf{Enc}_B$) is intentionally designed to harness this property. It allows any ciphertexts $\mathsf{ct}_A$ and $\mathsf{ct}_B$ to be arbitrarily paired and evaluated via a functional key to reveal their inner product. This unique property enables secure and decentralized aggregation of matrix products without losing LoRA's linear communication advantages. Furthermore, SecLoRA ensures round-isolated decryption to prevent temporal leakage without extra interaction. Evaluation shows that SecLoRA is practical for cross-silo deployments.

A private information retrieval (PIR) protocol enables a client to retrieve a bit from an $N$ bit database replicated among $k$ servers in such a way that each server learns no information about the retrieved bit. Modern PIR protocols with information-theoretic privacy (Efremenko, SICOMP, 2012; Dvir and Gopi, STOC, 2015; Ghasemi et al., STOC, 25) are based on matching vectors over a composite number $m$. To construct a PIR protocol from the matching vectors, these protocols use a decoding polynomial, a sparse polynomial that returns a non-zero value on 1 and returns zero on a certain set implied by the matching vectors.

Beimel et al. (CCC, 2012) abstracted the properties required by the transformation computed by the decoding polynomial, defining the notion of share conversion. In such a conversion, a set of parties is given shares of a secret in one secret-sharing scheme, and each party locally computes a new share (without any communication) such that the new shares are shares in a second secret-sharing scheme of a related secret. Beimel et al. showed that share conversion can replace the decoding polynomial in the PIR protocol of Efremenko and constructed a share conversion from the ring $\mathbb{Z}_6$ to the field $\mathbb{F}_{2^2}$. This share conversion cannot be computed by a decoding polynomial, as decoding polynomials convert shares from a ring $\mathbb{Z}_m$ to a finite field of characteristic $p$ such that $p$ does not divide $m$. Alon et al. (TCC, 2025) simplified and generalized the PIR protocols of Dvir and Gopi and Ghasemi et al., using share conversion; however, in this protocol, the share conversion is from a ring $\mathbb{Z}_m$ to a finite field of characteristic $p$ such that $p$ does not divide $m$.

In this paper, we study the power of share conversions. Our main result proves that if there is a $k$-party share conversion from a ring $\mathbb{Z}_m$ to a finite field of characteristic $p$ such that $p$ does not divide $m$, then there is a $k$ sparse decoding polynomial from a ring $\mathbb{Z}_m$ to a finite field of characteristic $p$. This result implies that using share conversion in the protocol of Alon et al. can only improve the communication complexity by a constant factor. In addition, we show that if there is a $k$-party share conversion from a ring $\mathbb{Z}_m$ to a finite field, where $m$ is a product of $r$ distinct primes, then $k\geq r+1$, i.e., the number of servers in the resulting PIR protocols using the appropriate matching vectors is at least $r+1$. A similar result was recently proved by Ghasemi and Kopparty (ITCS 26); our lower bound also applies to the case in which the characteristic of the field divides $m$.

Quantum cryptography is a rapidly developing area which leverages quantum information to accomplish classically impossible tasks. In many of these protocols, quantum states are used as long-term cryptographic keys, relying on the quantum no-cloning theorem to ensure that the keys cannot be copied by an adversary. Unfortunately, quantum state tend to decohere, and hence, persistent quantum memory is and will remain one of the most valuable and challenging resources for quantum computers. As such, it will be important to minimize the extent to which our protocols use persistent quantum memory.

In this work, we consider the case of one-shot signatures (OSS), and more general quantum signing tokens, important uncloneable primitives where quantum signing keys allow for signing a single message but not two. Very recently, the first OSS scheme was constructed unconditionally in a classical oracle model as well as in the standard model under cryptographic assumptions (Shmueli and Zhandry, CRYPTO 2025). We observe that the quantum memory required for these protocols is a large polynomial (in the security parameter).

The main contribution of this work is to significantly decrease the quantum secret key size, in some cases achieving the asymptotically optimal size. One of our schemes guarantees perfect correctness and the other one admits a parallel signing algorithm for long messages. We also achieve strong signature incompressibility, which implies a public-key quantum fire scheme (Çakan, Goyal and Shmueli, QCrypt 2025) with perfect correctness.

During the course of this work, we develop novel techniques for proving the security of cryptosystems using coset states, one of the main tools used in uncloneable cryptography.

Homomorphic Advanced Encryption Standard (AES) evaluation refers to evaluating the AES circuit with a fully homomorphic encryption (FHE)-encrypted secret key. Applications include in particular Transciphering, which converts AES-encrypted data into FHE ciphertexts without exposing the secret key.

Existing homomorphic AES evaluations show a clear separation between latency-oriented solutions and throughput-oriented solutions. CKKS-based methods exploit massive SIMD parallelism and focus on throughput by processing many AES blocks in parallel. They are hardly suitable for latency-critical settings. In contrast, TFHE-based methods process a small number of blocks efficiently. They are preferable for low-latency settings, but provide very limited throughput.

In this work, we show that AES-CKKS evaluation can achieve both interactive latency and high throughput. Our first variant is optimized for latency and decrypts a single AES block in only 34ms on an NVIDIA RTX-5090. This is more than 4× faster than recent TFHE-based state-of-the-art approaches. Our second variant is based on a new embedding of $\textrm{GF}(16)$, the finite field with 16 elements, into CKKS message space. It is optimized for throughput and processes up to 2048 AES blocks at once, achieving over 200KB/s throughput (a more than 3.1× improvement over the state-of-the-art CKKS-based approaches), while maintaining latency comparable to TFHE-based methods. To the best of our knowledge, this is the first AES-FHE evaluation algorithm combining good latency and throughput properties, bringing homomorphic outsourcing with AES within reach of real-time applications on constrained devices.

Our main ingredients are redundant structures that maximize SIMD utilization, improved algorithms for the SubBytes step (one of them being based on inversion in $\textrm{GF}(256)$ using CKKS), fusion of linear layers into bootstrapping, and carefully crafted FHE parameters.

We construct preimage sampleable function families on $q$-ary lattices [Gentry–Peikert–Vaikuntanathan, STOC'08] for which preimage sampling reduces to sampling uniform points of unitriangular lattices from simple polytopes: affine linear transforms on the $\ell_1$ and $\ell_\infty$ balls, and a scaled intersection thereof. We build the necessary samplers by adapting an algorithm of [Kannan–Vempala, STOC'97] and improving its analysis. This sampling requires only uniform bits and affine linear transforms on the uniform distribution on $[0,1]$. The collision resistance of these families relies on the short integer solutions problem [Ajtai, STOC'96] in various $\ell_p$ norms. Considering the Lee metric as the $\ell_1$ norm in $q$-ary lattices, we answer an open question to construct such families for the Lee metric [Hörmann–van Woerden, CRYPTO'24]. We also answer an open question of [Plançon--Prest, PKC'21] by sampling from polytopes with inradius smaller by a factor almost square root in the lattice rank. We provide a generic framework for constructing preimage sampleable function families from polytopes with sufficient conditions for realising it. While our parameters are worse than prior discrete Gaussian based constructions, such distributions are challenging from a physical security perspective.

Overbroad computer crime, intellectual property, and other laws are well known to create legal risks that can discourage essential research. Notable examples include the US Computer Fraud and Abuse Act and the UK Computer Misuse Act. Because such laws fail to distinguish malicious hacking from good-faith testing and research, researchers face serious legal risks for public-interest research activity like identifying software or hardware vulnerabilities or scraping data. Despite the research community's broad awareness of these risks, our understanding of their practical impacts is limited, as most of the community's knowledge comes from anecdotal evidence rather than systematic study.

We conduct the first qualitative study focused on researchers' lived experiences, to empirically document the *impacts of legal risks and threats* on research and researchers*, and *how researchers navigate legal risk situations*. Our study engages two participant groups: researchers with legal-risk experiences in the UK or the US ($N_R=36$), who discuss 130 projects and incidents spanning over three decades, and professionals that offer support to researchers navigating legal risks ($N_S=8$), who have collectively supported thousands of researchers. We thus provide an unprecedented big-picture view of researchers' experiences with legal risks. We synthesise actionable strategies for researchers, and our findings provide evidence to support policy reform.

We show new constructions for pseudorandom quantum states (PRS) and pseudorandom function-like quantum state (PRFS) generators satisfying scalability, which means the security parameter can be much larger than the number of qubits, quantum accessibility, which means the adversary can provide quantum input, and adaptivity, which means the adversary can query it adaptively. We present an isometric procedure to prepare quantum states that can be arbitrarily random (i.e., the trace distance from the Haar-random state can be arbitrarily small for the true random case, or the distinguishing advantage can be arbitrarily small for the pseudorandom case). This naturally gives the first construction for scalable, quantum-accessible, and adaptive PRFS assuming quantum-secure one-way functions. Compared to prior PRFS works, we use a stronger definition of quantum accessibility in which the adversary can be ancilla-assisted, i.e., the input state may not be pure and could be entangled with other quantum registers. Thus, our result also gives the first (fully) quantum-accessible PRFS. Our PRFS construction implies various primitives, including long-input PRFS, short-input PRFS, short-output PRFS, non-adaptive PRFS, and classically-accessible adaptive PRFS. This new construction may be helpful in simplifying the microcrypt zoo.

This work proposes StakeNote, a distributed ledger protocol that combines Proof-of-Stake (PoS) with privacy and anonymity preserving payments. The protocol combines Ouroboros Praos, a provably secure PoS protocol, with CryptoNote, a privacy-preserving payment system based on ring signatures which has been widely used in practice. We prove that StakeNote inherits the security guarantees of Ouroboros Praos and the privacy guarantees of CryptoNote and we demonstrate its practicality via a proof of concept implementation, where block creation requires less than 25 ms and eligibility proofs are approx. 3 KB for anonymity sets of size 16. Finally, we discuss heuristic enhancements that potentially increase privacy and enable dynamic participation.

End-to-end deep-learning side-channel attacks on public-key implementations have recently become possible even for million-sample traces. However, existing methods require large computational resources and extract only partial key shares, which means that dedicated post-processing is required to turn detected leakage into demonstrations of successful key recovery attacks. We present an end-to-end sequence-to-sequence prediction approach to recover complete 256-bit key shares from single raw traces on the SCAAML ECC datasets recently studied by Bursztein et al (TCHES 2024).

Our solution combines aggressive trace compression for dimensionality reduction with a 1-D U-Net trained using Connectionist Temporal Classification loss. The key idea is to decouple detecting leakage from mapping each leakage site to the correct part of the secret: the network outputs an annotated map of the trace marking likely leakage sites, and a greedy decoder reconstructs the ordered key bits from that map. Using synthetic tasks, we show that this division of labor circumvents a fundamental problem that causes neural network architectures and training methods commonly used in side-channel analysis to struggle with massive multi-target or misaligned extraction tasks.

As a result, we are able to train a single extractor that achieves high accuracy on all four SCAAML ECC datasets in a single training run that takes minutes on a single GPU. The resulting extractors are robust, essentially maintaining their performance under large misalignment (we empirically tested rotations up to \(61\%\) of trace length), while degrading gracefully under a variety of trace corruptions, and even time reversal. They transfer across key shares and datasets with little degradation and no retraining. The U-Net outputs also yield prediction maps that localize leakage along the trace prior to decoding.

Cryptographic signatures play an integral part in ensuring authenticity and integrity in digital systems. Their security crucially relies on the secrecy of the signing key, since knowledge of this key enables an adversary to generate valid signatures on any message. Once a signing key is compromised, the standard countermeasure is to revoke the corresponding public key and to invalidate all signatures produced for this key. However, with this approach even legitimate signatures created by the honest signer would retroactively lose their validity. In this work, we initiate the formal study of a new approach - Signatures with Post-Compromise Accountability (SPCA) - which provides security guarantees even after the secret key was compromised. This notion effectively introduces a grace period for the legitimate key owner, during which the validity of honestly generated signatures is preserved despite the adversary’s knowledge of the secret key. We formally define SPCA and its security guarantees, and present two constructions achieving this notion. Our first construction generalizes the signature-in-signature approach of Błaśkiewicz et al. (ESORICS '21), where an inner signature is embedded into the randomness of an outer signature. This construction, however, requires revealing the signing secret key during revalidation. Our second construction overcomes this limitation by enabling revalidation without disclosing the secret key, yielding stronger security guarantees.

Cryptographic accumulators based on groups of unknown order (GUO) provide constant-size set membership proofs. For security purposes, existing works require first encoding set elements via division-intractable (DI) hash functions, typically instantiated as random oracles that destroy any algebraic structure. This confines GUO-based accumulators to a purely set-membership role, making them "incompatible" with various existing cryptographic proof techniques over committed integers in the same groups as the GUO, such as constant-size proofs of exponentiation and modular exponent relations. We introduce the notion of morphic accumulators, which replaces the DI hash with a discrete logarithm encoding $H_g(x) = g^x$, mapping set elements to a group before accumulation. We prove, under a variant of the subset product assumption in the generic group model, that this encoding is inherently division intractable, achieving the same security guarantee as random-oracle DI hashes, while simultaneously being a group homomorphism: accumulated elements retain their group-algebraic relationships. This resolves a fundamental tension between compact representation and algebraic structure: the accumulator serves simultaneously as a binding commitment to a set and as a substrate for homomorphic computation over its elements.

Morphic accumulators yield asymptotically optimal constructions across multiple domains: range proofs with $O(n)$ prover time, $O(1)$ proof size, $O(1)$ verification with transparent setups (the first scheme to simultaneously achieve these optimal bounds); polynomial commitments with $O(n)$ prover and $O(1)$ proof size, resolving the cubic bottleneck in prior constant-proof-size GUO-based schemes; and the first linkable ring signatures with $O(1)$ signature size, transparent setup, $O(n)$ offline signing and $O(1)$ online signing.

In the line of the SPDZ protocol for secure multi-party computation, the generation of Beaver triples is the most expensive task. Silentium (Rieder, PrivCryp 25) is the implementation of a Pseudorandom Correlation Generator (PCG) for Beaver triples (Boyle et al., Crypto 20). PCGs focus on low-communication costs., e.g. their PCG reduces the communication by one order of magnitude compared to protocols in MP-SPDZ. Silentium is an implementation of their PCG, achieving similar running times than MP-SPDZ. We make three theoretical contributions to Silentium, including an implementation. First, we make a practical proposal how to generate Beaver triples over binary fields F2λ, which extends the previous setting over prime fields. For this, we propose a suitable instantiation of the Number Theoretic Transform. Second, we show how to use the binary triples to construct what we call a Beaver triple expansion scheme, that is we construct a scheme that expands a small batch of Beaver triples into a large batch of Beaver triples, in the sense of recently established oblivious transfer extension schemes. This feature enables an efficient preprocessing stage for the PCG, closing a practical issue of Silentium. Finally, we provide details about the Silentium implementation, by clearing a technical bug in the initial theoretical protocol description.

Functional bootstrapping has made discrete computation practical in the Cheon-Kim-Kim-Song (CKKS) scheme, but it fuses four distinct tasks - lookup table (LUT) evaluation, modular reduction, noise cleaning, and ciphertext refreshing - into a single rigid pipeline. As a consequence, a generic LUT over an alphabet of size $t$ costs multiplicative depth proportional to $\log_2 t$ and consumes a large share of the modulus budget during a fixed bootstrapping procedure, invoked each time a LUT evaluation or modular reduction is needed. We show that this pipeline can be unbundled by changing the representation, rather than optimizing the bootstrapping, through block encodings. A finite-alphabet value is carried across several CKKS slots whose coordinates form a basis of functions on the alphabet, typically the characters of a finite abelian group. In such a basis, every LUT is an affine plaintext map evaluated in a single multiplicative level, with depth independent of $t$. Modular reduction comes for free: a block encoding cannot represent anything but a residue, so arithmetic modulo $t$ is native. Because the encoded values lie on the unit circle, noise growth is independent of the alphabet size $t$. In the worst case, it matches the noise growth of standard discrete CKKS on the smallest alphabet $\mathbb Z_2$, and in more typical workloads it is linear in the number of operations, exponentially better than discrete CKKS at every $t > 2$. Noise cleaning becomes a constant-depth procedure of at most four levels, because the alphabet-dependent part is an LUT and only a fixed-degree-3 smoothstep is nonlinear. Finally, since LUTs are no longer part of the bootstrapping, refreshing reverts to its classical role as a maintenance operation invoked only to regain multiplicative depth. Any CKKS bootstrapping can be used, rather than a constrained and expensive pipeline. We instantiate the framework with several block encodings that make modular addition, modular multiplication, xor or min/max possible with a single CKKS multiplication. We use them to build CRT arithmetic over large composite moduli, and finite-state prefix scans for radix addition and subtraction in depth $4 + \lceil\log_2 d\rceil$ and for equality and comparison in depth $3 + \lceil\log_2 d\rceil$ for $d$ radix digits. For example, a 256-bit CRT modular addition or multiplication consumes a single multiplicative level and has a latency of 4.7 ms on a single thread.

We present a post-quantum commitment scheme based on kernel-tagged punctured Richelot isogeny walks on superspecial genus-2 Jacobians. The puncturing rule skips every step landing in the product locus, detected by I10 = 0, so honest executions remain in the Jacobian locus and avoid the entry point of known product-locus attacks. Each opening is encoded as a deterministic non-backtracking walk together with a kernel tag recording its action on a small public auxiliary torsion basis. The tag is verified as part of the opening and is kept explicit throughout the security analysis. In particular, scalar-related collisions force equality of the ordered kernel sequence and hence equality of the tag, so any nontrivial binding attack yields a short non-scalar endomorphism. Using spectral bounds for the Richelot graph, we show that puncturing preserves rapid mixing for logarithmic walk lengths, which yields statistical hiding for the tagged punctured scheme. We therefore reduce binding to the Short Richelot Endomorphism Problem (SREP), relate SREP to the One-Endomorphism Problem and, under a standard KLPT2 -style heuristic, to the endomorphism-ring problem. A SageMath prototype based on (2, 2)-Kummer isogenies indicates practical performance at standard security levels.

Finding a nontrivial endomorphism of a given supersingular elliptic curve is a hardness assumption of isogeny-based cryptography. We prove the reduction from it to the problem of finding a splitting of a given principally polarized abelian surface. By using this new reduction, we also prove the heuristic equivalence of the splitting problem with a degree restriction and the endomorphism ring problem in dimension two.

Batched encryption (BE) has emerged as a novel public-key cryptographic paradigm that enables the efficient decryption of a designated batch of $B$ ciphertexts simultaneously. By incorporating threshold decryption capabilities into this framework, batched threshold encryption (BTE) further decentralizes the decryption process. While both BE and BTE serve as highly effective solutions for mitigating Miner Extractable Value (MEV) attacks in blockchain networks by providing robust mempool privacy, ciphertext integrity, and communication efficiency, they still suffer from heavy computational overhead during the ciphertext decryption phase.

In this paper, we address this computational bottleneck by introducing a novel framework that delegates the heavy decryption workloads to an untrusted cloud server while enabling verifiability of the outsourced computations. To achieve this, we first propose an outsourced batched identity-based encryption (O-BIBE) scheme by integrating outsourcing functionalities into the conventional BIBE paradigm, accompanied by a rigorous security proof. We then construct a verifiable outsourced batched encryption (VOBE) scheme by strategically combining O-BIBE with other core cryptographic building blocks and formally prove its security.

To eliminate the single point of failure and enhance threshold resiliency, we extend our framework to the threshold setting by developing an outsourced threshold batched identity-based encryption (O-TBIBE) scheme. Building upon this, we propose a verifiable outsourced batched threshold encryption (VOBTE) scheme, which successfully achieves decentralized threshold resilience. Our proposed VOBE and VOBTE schemes are the first to concurrently guarantee ciphertext integrity and mempool privacy against sophisticated blockchain attacks, while significantly reducing decryption costs via efficient and verifiable outsourcing.

We present Grand Danois, a new post-quantum multilinear polynomial commitment scheme from lattices for polynomials over $\mathbb{F}_q$ that achieves polylogarithmic $O(\lambda \ell)$ verification complexity and proof sizes. We build on the general approach introduced in Hachi (ePrint 2026/156) with two key changes. First, we switch to the vanishing Short Integer Solution (vSIS) assumption to obtain structured public parameters for our commitment scheme and utilize this structure to design an adapted sumcheck protocol amenable to succinct verification. Second, we modify the quadratic relation used in Hachi and Greyhound (CRYPTO 2024) so that it becomes compatible with proving norm bounds using Johnson-Lindenstrauss projections. This is achieved through an adaptation of the structured projection strategy introduced in RoK and Roll (ASIACRYPT 2025). This has the benefit for communication complexity in that proving norm bounds and correct polynomial evaluation are integrated into a single protocol, reducing the number of commitments sent by the prover. Furthermore, we impose additional structure on our random projections to reduce the witness size even more aggressively during each round of recursion without sacrificing verification complexity. Under the vSIS assumption, our construction yields an estimated proof size of roughly $80-90$ KB for $2^{32}$-size polynomial evaluations.

Blockchain mempool transparency fuels Maximal Extractable Value (MEV), where attackers can front-run, back-run, and reorder transactions as soon as they appear. Encrypted mempools aim to delay the release of information until block commitment, yet nearly all existing designs rely on a trusted decryption committee. This creates two structural problems. First, committee members hold decryption material by design, so a colluding threshold can reconstruct the decryption key and learn transactions before block commitment. Second, once such a committee becomes malicious, honest parties have no easy in-protocol way to recover: restoring privacy for future epochs requires an external intervention such as a hard fork that replaces the committee and rotates the long-lived cryptographic material.

In this work, we ask whether encrypted mempools can instead use proof-of-work to realize an open and recoverable decryption committee. We then introduce DecryptChain, a permissionless proof-of-work encrypted mempool in which decryption authority is not assigned to persistent identities or long-lived key shares. Instead, decryption is continuously re-contested through public computational work. Even if an adversary successfully breaches one epoch, it gains no reusable secret for future epochs; honest parties can always re-enter and recover the decryption process by contributing sufficient work. DecryptChain decouples block production from decryption, enabling it to operate as a Layer-2 timely decryption service on any underlying blockchain while preserving eventual decryption for committed on-chain encrypted transactions.

The communication complexity of unconditionally Secure Multi-Party Computation (MPC) protocols has been studied by a series of works in the honest-majority setting. For evaluating an arbitrary Boolean circuit, the state-of-the-art MPC protocol by Goyal et al. (Crypto 2021 and Crypto 2022) achieves the total communication cost of $O(\log n)$ bits per gate, where $n$ is the number of parties. In this work, we present the first unconditional MPC protocol for any Boolean circuit with communication of $O(1)$ bits per gate. We first construct an unconditionally secure protocol in the presence of semi-honest adversaries, and then strengthen it to guarantee security against malicious adversaries with the same communication efficiency.

This paper proposes a systematic approach to compute cryptanalytic properties of arbitrary Mealy machines or S-functions. Based on the geometric approach to cryptanalysis, we provide a uniform formula for any cryptanalytic property of such a function, as long as the property is compatible with the way its input and output are split into chunks. Examples include linear, (quasi) differential, (ultrametric) integral, differential-linear, and boomerang properties. To illustrate our results, we compute these properties for several important examples, including modular additions, the Chi- and ChiChi-functions, and the SHA-1 step function. As proof-of-concept applications, we construct a boomerang distinguisher for the Subterranean permutation, and show how to compute the correlations of conditional linear approximations in partitioning-based differential-linear attacks more accurately. Our results also lead to a new approach to compute the algebraic normal form of the inverse of the Chi-function.