International Association
for Cryptologic Research

Large Language Models (LLMs) are frontier deep learning systems that have achieved remarkable success across a wide range of AI services. However, their substantial computational and memory requirements make them difficult to deploy and run on local hardware. Due to these resource requirements, users often rely on untrusted cloud infrastructure providers to perform model inference. However, outsourcing introduces the challenge of verifying that the returned output is the genuine result of the specified model. In this work, we present DeepProve, the first system to enable efficient end-to-end verification of full LLM inference (i.e., for all generated tokens of a prompt) on untrusted cloud servers using zero-knowledge proofs (ZKPs). In contrast, prior work either provides only a proof-of-concept partial implementation for a single token (zkGPT, USENIX'25), or focuses exclusively on specific components of the inference pipeline, such as Softmax (zkLLM, CCS'24).

DeepProve achieves end-to-end verification by certifying the correctness of the output sequence rather than encoding the expensive inference computation in-circuit, an approach that would require either circuit size quadratic in the sequence length or costly in-circuit modelling of RAM operations. The core building blocks of DeepProve are sum-check protocol and lookup arguments, which enable efficient proof of correctness of all operators needed for GPT-2 and Gemma 3, such as multi-head attention and layer normalization for GPT-2, and grouped-query attention, root mean square normalization, and rotary positional embeddings for Gemma 3. Our evaluation shows that DeepProve can prove inference of GPT-2 and Gemma 3 at approximately 174 and 86 tokens per minute, respectively, which is 20-60 faster than the state of the art, without any significant loss in accuracy. Verification takes only 1 to 3.7 seconds. By distributing proof computation across multiple nodes, DeepProve can further improve the prover time while reducing the memory requirements for individual machines. With distributed proving, DeepProve can scale the throughput to 1855 tokens per minute. Our work represents the first full system for end-to-end LLM inference verification, thus paving the way for secure and trustworthy AI services.

Revocable and linkable ring signatures ($\mathsf{RLRS}$) were introduced by Zhang et al. (Inscrypt 2019), where a revocation authority (RA) can mandatory revoke the anonymity of the real signer in cases of dispute. However, existing $\mathsf{RLRS}$ constructions face practical challenges. First, the reliance on Public Key Infrastructure (PKI) incurs a substantial burden of certificate management. Second, current schemes suffer from computational or communication overheads that scale linearly with the ring size, hindering their application in large-scale scenarios. Third, prior schemes often assume a fully trusted RA, leaving honest users vulnerable to being framed by a compromised or malicious authority. In this paper, we formalize the notion of Identity-Based Revocable and Linkable Ring Signature ($\mathsf{IB\text{-}RLRS}$), which extends the $\mathsf{RLRS}$ to the ID-based setting and bypasses the need for cumbersome PKI management. Specifically, $\mathsf{IB\text{-}RLRS}$ ensures that the real signer's identity can always be extracted, and that a malicious RA cannot frame honest users who did not participate in the signature generation. Based on the IEEE P1363 standard, a secure $\mathsf{IB\text{-}RLRS}$ scheme is proposed. Specifically, we design a novel ring signature construction featuring logarithmic signature size that requires no trusted setup, and formally prove its security in the random oracle model.

Abstract. The TWEAKEY framework provides a generic construction for designing tweakable block ciphers. Prominent instances are Deoxys-BC and SKINNY, which have been standardized by ISO/IEC. In this paper, we analyze the tweakey schedules of these ciphers and identify strong dependencies between certain subtweakeys, which we call strong key bridges. We then exploit these dependencies in rectangle attacks under the related-tweakey setting. Moreover, we develop a comprehensive constraint programming model to search for rectangle attacks. Our model not only unifies the distinguisher and the key-recovery part while permitting arbitrary key-guessing strategies, but also integrates three new components, i.e., the state-test technique, explicit last-step computation, and the strong key bridges. As a result, we obtain significantly improved cryptanalytic results on both Deoxys-BC and SKINNY. For Deoxys-BC-384 and Deoxys-BC-256, we reduce the time complexity by a factor of $2^{40}$ and $2^{32}$, respectively, yielding the best attacks to date. Moreover, we extend the longest existing attacks on the authenticated encryption schemes Deoxys-I-256 and Deoxys-I-128 by one round. For SKINNY, we improve upon prior best attacks by one more round for SKINNY-128-384 and by two more rounds for SKINNY-64-192.

Public permissionless ledgers are append-only by design, yet are increasingly asked to host data subject to statutory erasure obligations (GDPR Art. 17, EDPB Guidelines 02/2025). The prevailing cryptographic response, rewriting the chain via chameleon hashes and redactable-blockchain constructions, is structurally unavailable on Bitcoin and Ethereum mainnet. This SoK systematizes the orthogonal application-layer design space, in which the ledger is left untouched and erasure is achieved by destroying the key material required to read the on-chain bytes. We organise the literature along two axes, data locus (on-chain ciphertext, anchor-only, or hybrid) and key custody (single-custodian, threshold, time-lock, or witness-encryption), into a twelve-cell grid with canonical constructions from 2006 to 2026. We extend standard IND-CCA security with a key-lifecycle adversary model parametrised by HSM side-channel leakage, coercion fraction over the custody committee, and algorithmic-break horizon, and establish an equivalence proposition between the resulting Destruction-IND notion and the EDPB "render unrecoverable" criterion.

Evaluating seven reference architectures against eleven engineering, cryptographic, and regulatory criteria, including a production RegTech engine processing more than 10^4 events per second, we find that the literature has concentrated on chain-rewriting while the application-layer branch lacks a common threat-model vocabulary. We close this gap and identify five open problems: post-quantum equivalence, forward-secure destruction under leakage, decentralised lifecycle audit (with a multi-custody extension covering encrypted mempools), composition with ZKP selective disclosure, and erasure under multi-regime retention conflict.

End-to-end encryption (E2EE) provides strong confidentiality guarantees to users by preventing service providers from accessing their data. At the same time, it introduces new operational challenges, most notably the restoration of an E2EE-protected backup on a new device after loss of the original device. In recent years, major instant messengers have deployed increasingly sophisticated key-retrieval schemes for encrypted backups, ranging from simple recovery codes to designs that depend on trusted hardware or advanced cryptographic primitives such as oblivious pseudorandom functions (OPRFs), threshold OPRFs, and secret sharing. While recent work has largely emphasized stronger cryptographic guarantees, the practical costs and trust assumptions implied by hardware dependencies and complex primitives are often not made explicit.

We present a structured framework for the comparative evaluation of E2EE backup key-retrieval schemes across usability, deployability, and security properties, and apply it to simple recovery codes, real-world deployments, and recent proposals from the literature, explicitly distinguishing recovery that depends on user-chosen low-entropy secrets from recovery using high-entropy material. Our evaluation shows that schemes whose recovery ultimately relies on user-chosen low-entropy secrets can provide substantially weaker resistance against large-scale adversaries than an approach using simple recovery codes. Even when assuming high-entropy recovery secrets, many recovery protocol designs introduce additional trust assumptions and potential single points of failure (e.g., compromise of specialized components or trusted setups) that can amplify the impact of large-scale adversaries. Overall, our results highlight that added system complexity often trades off against limited security gains under realistic threat models, motivating clearer accounting of cost and distribution of trust in future designs.

Stateful hash-based signature schemes such as XMSS and LMS are increasingly important in post-quantum cryptographic deployments, yet their verification paths remain difficult to fuzz effectively because key generation is substantially more expensive than verification. This paper presents a structured libFuzzer methodology for testing stateful hash-based signature verification in liboqs. The proposed approach uses pre-computed Known Answer Test (KAT) vectors to initialise valid public key, signature, and message buffers once during harness setup, enabling high-throughput mutation of verification inputs without repeated key generation.

Using this methodology, we implemented and upstreamed two fuzz harnesses targeting XMSS/XMSSMT and LMS/HSS-LMS verification paths in liboqs. A ten-minute AddressSanitizer-enabled fuzzing campaign identified an OID-confusion heap-buffer-overflow in xmssmt_core_sign_open, assigned CVE-2026-46344. The flaw occurs when attacker-controlled public-key OID bytes derive a parameter set whose expected signature length exceeds the caller-allocated buffer, resulting in an out-of-bounds read during signature parsing. The issue was reproduced using a five-byte minimal crashing input and a standalone C reproducer, disclosed through the Open Quantum Safe coordinated disclosure process, and fixed in liboqs 0.16.0.

We further identify a mutation-distribution bias affecting short-input LMS fuzzing and resolve it using a round-robin field-rotation strategy. The resulting framework provides a reproducible and upstream-integrated methodology for adversarial testing of stateful post-quantum signature verification implementations.

We present a Byzantine agreement protocol to address the inefficiencies inherent in multi-valued Byzantine agreement protocols, i.e., a version of the Byzantine agreement protocol where every party broadcasts its request, and at the end of the protocol, every party agrees on one of the party’s requests. The protocol we present is a validated asynchronous Byzantine agreement protocol, i.e., a party’s request must be validated by some external validity property before it is proposed for agreement. Differently from most of the MVBA protocols, we allow only a subset of total parties to broadcast their requests instead of all, and we make the subset selection stochastic each time the parties choose to broadcast a new set of requests. Then, at the time of the agreement, we choose a party from the selected subset, and the parties reach an agreement on the selected party’s broadcast. Extensive theoretical analysis shows that this approach can produce efficient output regarding messages and computation overhead, but the protocol is time-consuming.

Event date: 11 April to 15 April 2027
Submission deadline: 17 September 2026
Notification: 18 January 2027

Event date: 2 March to 13 March 2026

Event date: 27 July to 7 August 2026

Event date: 6 July to 7 July 2026

Event date: 22 August to 28 August 2026

Event date: 3 December to 4 December 2026
Submission deadline: 25 November 2026
Notification: 29 July 2026

Minimizing round complexity is a central goal in secure Multi-Party Computation (MPC), particularly for deployment on high-latency networks. While constant-round protocols with concrete efficiency have been constructed, they are typically designed for Boolean circuits and each gate incurs a bandwidth cost linear in the security parameter. Moreover, for arithmetic-heavy applications such as privacy-preserving machine learning and statistical analysis, compiling arithmetic operations into Boolean gates incurs another substantial overhead in circuit size and communication. Conversely, existing arithmetic MPC protocols, such as SPDZ, require interaction rounds proportional to the circuit depth, imposing significant latency.

In this work, we bridge this gap by presenting the first concretely-efficient maliciously-secure MPC protocol that achieves both constant-round and constant-rate communication, where the rate is defined as the bandwidth cost per party divided by the number of gates and the size of the values each gate operates on. Our protocol computes over bounded integers and is secure against a static, malicious adversary corrupting up to $n-1$ parties. The protocol is built upon the arithmetic garbling framework of Ball et al. (Eurocrypt 2023) and follows the BMR template, assuming the Decisional Composite Residuosity for the garbling phase and Learning Parity with Noise for preprocessing.

We evaluate our protocol on matrix-vector multiplication, a fundamental operation for data analysis. For standard computation parameters, we reduce communication bandwidth by $101\times$ to $247\times$ and improves end-to-end runtime by $4.4\times$ to $10.7\times$ compared to state-of-the-art constant-round Boolean MPC baselines, even when accounting for the overhead of a full bit-decomposition on the output vector.

Designing a secure symmetric-key cipher over a vector space over a field $\mathbb F_{p^n}^t$ is well known and understood by the cryptographic community. Even if the attacks are continuously improving, our current understanding regarding the design and security of the majority of the symmetric-key primitives has not fundamentally changed in the last 20 years.

How does this picture change when we move to an integer ring $\mathbb Z_{p^n}^t$? Although the question is easy to state, it turns out to be far harder to answer. Indeed, there is a significant difference between the arithmetics of $\mathbb F_{p^n}^t$ and $\mathbb Z_{p^n}^t$ and attack vectors do not apply/translate directly between the two. As a case in point, a few ciphers have already been designed over integer rings, yet their initial versions have already been broken.

In this paper, we lay the foundations for a more rigorous approach to designing ciphers over integer rings, noting that this is not only of theoretical interest, but also has concrete applications. We analyze how existing statistical and algebraic attacks will behave for these ciphers and also present new attacks that take into account that not all functions over integer rings admit a polynomial representation. Based on this, we discuss possible design strategies, in which we analyze the security effect of having/not having polynomial S-boxes. In particular, we introduce new properties for the non-polynomial S-boxes that measure their resistance against the attacks presented in this paper. Finally, we discuss how to design such non-polynomial S-boxes, presenting two concrete constructions, and one based on the "digit manipulation".

Few-time signatures cap how many signatures a signer can safely issue. Jevil is, to our knowledge, the first post-quantum and transparent (setup-free) few-time signature scheme with a sharp key-recovery cliff: its cap is enforced by a single sharp threshold rather than a slow slope. Signatures one through $n^{\star}$ are existentially unforgeable at approximately $124$-bit classical security; at the $(n^{\star}+1)$-th the entire secret polynomial becomes publicly recoverable. The cap is founded on a secret polynomial together with the degree-binding of a polynomial commitment, and is intrinsic to any accepted public key: even a malicious signer who chooses $\mathsf{pk}$ adversarially cannot construct one that lets them keep signing past the cliff without the same polynomial becoming publicly recoverable, achieving catastrophic failure as a key design requirement. All prior post-quantum few-time schemes (HORS, FORS, PORS, HORSIC$^{+}$, eBiBa, Syrga$_2$) degrade softly as $(nK/T)^K$; cliff-style behaviour was previously confined either to one-time-only Schnorr/ECDSA nonce reuse (not post-quantum, not designed) or to polynomial-witness constructions that depend on KZG/IPA commitments (not post-quantum, not transparent).

Concretely, Jevil provides $\sim 68$-byte public keys, $32$-byte secret keys, and $40$-$500$KB signatures across signing budgets $n^{\star} \in \{1, 3, 7, 15, \ldots, 2^{14} - 1\}$, the limit imposed by the $2$-adicity of the working field. All primitives are believed to be post-quantum.

We propose a CKKS-based technique for evaluating arithmetic over finite fields F_{p^r} with small characteristic p under homomorphic encryption. The core of our approach is a pair of complementary ciphertext representations. In the so-called spectral encoding, ciphertext addition and multiplication realize addition and multiplication in the field F_{p^r}. In another encoding, coefficient encoding, the same operations act as slotwise addition and multiplication in the slot algebra (F_p)^r. We show that one can switch homomorphically between these encodings at cost linear in r, and that F_p-linear maps, such as taking p-th powers in F_{p^r}, can be folded into these switches or applied directly in either representation. We complement the construction with theoretical and practical correctness-management techniques. To support unbounded computations, we integrate our framework with existing CKKS bootstrapping techniques and benchmark it against BGV-based implementations of F_{p^r}-arithmetic, a natural baseline for high-throughput finite-field computation. Across the fields we tested, this yields speedups ranging from 1.7x to 178x in amortized multiplication time when bootstrapping is taken into account. The gains are parameter-dependent: roughly speaking, our advantage over BGV increases as the characteristic p becomes smaller and the extension degree r becomes larger.

Techniques to randomly select sets of anonymous parties are ubiquitous in efficient and adaptively secure consensus protocols, as well as in Multi-Party Computation in the YOSO model, where each round is executed by a different random anonymous committee. Anonymous committee selection aims at randomly selecting a set of $n$ parties (the committee), where at most $t$ parties are corrupted (except with negligible probability), drawing from a population of $N \gg n$ parties with at most $T$ corrupted parties. Additionally, each party knows (and can prove) if they belong to the committee, but ignores other members' identities. A very common and efficient instantiation of anonymous committee selection is to select parties according to a VRF output, this however, leads to committees of probabilistic size ($n$ behaves as a Binomial random variable). Despite wide adoption, only Blum et al. (CCS23) provides an analysis of VRF-based probabilistic anonymous committee selection that estimates the size of committees. This analysis relies on lose bounds (Chernoff) and approximations (Poisson). In this work, we revisit Blum et al.'s estimates and derive accurate closed-form formulas (based on a tight Binomial approximation), as well as an efficient high-precision library called Tail-Hammer for computing exact parameters. Notably, Tail-Hammer identifies smaller committee sizes (approximately -25% on average) than Blum et al. (CCS23) for the same security level, leading to improved efficiency in protocols relying on random committee selection, also when anonymity is not needed. Our analysis applies to committee selection techniques that employ unbiased (uniformly random), or bounded-bias randomness, to both synchronous and asynchronous communication settings, and it can account for inactive parties. As a new application, we present a verifiable consistent broadcast protocol that leverages quorums in anonymous committees to achieve efficiency without requiring threshold signatures.

Publicly verifiable secret sharing (PVSS) is a fundamental primitive in threshold cryptography that allows a dealer to share a secret $S$ among a set of $n$ parties via a publicly verifiable transcript. Any subset of $t+1$ parties can then use their individual shares to reconstruct the full secret $S$, whereas $t$ or fewer shares give no information about $S$. As such, the secret $S$ remains hidden from an adversary that corrupts up to $t$ parties. Recently, Bacho and Loss (CCS 2023) gave the first proof of any PVSS scheme under an adaptive adversary. However, their security proof relies on strong and non-standard assumptions such as the algebraic group model (AGM) and the hardness of the one-more discrete logarithm (OMDL) problem. In particular, any protocol (e.g., distributed randomness beacon or distributed key generation) that makes use of a PVSS scheme either inherits these limitations or is not provably adaptively secure.

In this work, we present for the first time an adaptively secure PVSS scheme from well-established assumptions. In more detail, we provide two PVSS schemes with different properties. Our first scheme works over any pairing-free cyclic group and its security relies on the decisional Diffie-Hellman (DDH) assumption. Our second scheme works over an asymmetric pairing group, its security relies on the DDH and the co-computational Diffie-Hellman (co-CDH) assumption, and has the particularly valuable feature of aggregatability, which allows the aggregation of multiple PVSS transcripts into a single transcript while preserving verifiability. Notably, both our schemes are highly efficient, non-interactive, and work in the established plain public key model. These properties along with their provable adaptive security make them suitable candidates as building block in higher-level distributed protocols that aim to minimize communication.

VOLE-in-the-Head (VOLEitH) is one of the most promising frameworks to design post-quantum digital signatures based on symmetric primitives. However, all existing symmetric primitives do not capture the specialized characteristics of the VOLEitH framework and are not VOLEitH-friendly, leaving room for improving the efficiency of VOLEitH-based signatures. In this paper, we propose a VOLEitH-friendly symmetric primitive called Lynx, which is optimal in terms of the number of required VOLE correlations that directly determines the efficiency of VOLEitH-based signature schemes. In particular, Lynx adopts a multi-branch structure featuring a new truncation function: (a) nonlinear components are customized to minimize the witness length and polynomial degree, as well as the number of finite-field multiplications; (b) linear layers are strategically interleaved to strengthen security. The security of Lynx is rigorously validated by covering all possible attacks in the presence of both classical and quantum adversaries. Built upon Lynx, we design a post-quantum signature scheme, Lynxer, in the VOLEitH framework, which is shorter and faster than all known post-quantum signature schemes from symmetric primitives. According to our experimental results, compared to the state-of-the-art symmetric-based signature schemes in the same setting, i.e., Rainier (CCS’22), AIMer (CCS’23) and FAESTv2 (Crypto’25), our signature scheme Lynxer reduces the “public-key size + signature size” by 25% ∼ 51%, and improves the signing (resp., verification) time up to 90.6% (resp., 89.3%).