International Association
for Cryptologic Research

Verifiable random functions (Micali et al., FOCS'99) allow a key-pair holder to verifiably evaluate a pseudorandom function under that particular key pair. These primitives enable fair and verifiable pseudorandom lotteries, essential in proof-of-stake blockchains such as Algorand and Cardano, and are being used to secure billions of dollars of capital. As a result, there is an ongoing IRTF effort to standardize VRFs, with a proposed ECVRF based on elliptic-curve cryptography appearing as the most promising candidate.

In this work, towards understanding the general security of VRFs and in particular the ECVRF construction, we provide an ideal functionality in the Universal Composability (UC) framework (Canetti, FOCS'01) that captures VRF security, and show that ECVRF UC-realizes this functionality.

We further show how the range of a VRF can generically be extended in a modular fashion based on the above functionality. This observation is particularly useful for protocols such as Ouroboros since it allows to reduce the number of VRF evaluations (per slot) and VRF verifications (per block) from two to one at the price of additional (but much faster) hash-function evaluations.

Finally, we study batch verification in the context of VRFs. We provide a UC-functionality capturing a VRF with batch-verification capability, and propose modifications to ECVRF that allow for this feature. We again prove that our proposal UC-realizes the desired functionality. We provide a performance analysis showing that verification can yield a factor-two speedup for batches with 1024 proofs, at the cost of increasing the proof size from 80 to 128 bytes.

Many online applications, such as online file backup services, support the sharing of indexed data between a set of devices. These systems may offer client-side encryption of the data, so that the stored data is inaccessible to the online host. A potentially desirable goal in this setting would be to protect not just the contents of the backed-up files, but also their identifiers. However, as these identifiers are typically used for indexing, a deterministic consistent mapping across devices is necessary. Additionally, in a multi-device setting, it may be desirable to maintain an ability to revoke a device’s access—e.g. through rotating encryption keys for new data.

We present a new primitive, called the Oblivious Revocable Function (ORF), which operates in the above setting and allows identifiers to be obliviously mapped to a consistent value across multiple devices, while enabling the server to permanently remove an individual device’s ability to map values. This permits a stronger threat model against metadata, in which metadata cannot be derived from identifiers by a revoked device colluding with the service provider, so long as the service provider was honest at the instant of revocation. We describe a simple Diffie- Hellman-based construction that achieves ORFs and provide a proof of security under the UC framework.

We present experimental findings on the decoding failure rate (DFR) of BIKE, a fourth-round candidate in the NIST Post-Quantum Standardization process, at the 20-bit security level. We select parameters according to BIKE design principles and conduct a series of experiments. We directly compute the average DFR on a range of BIKE block sizes and identify both the waterfall and error floor regions of the DFR curve. We then study the influence on the average DFR of three sets $\mathcal{C}$, $\mathcal{N}$, and $2\mathcal{N}$ of near-codewords --- vectors of low weight that induce syndromes of low weight --- defined by Vasseur in 2021. We find that error vectors leading to decoding failures have small maximum support intersection with elements of these sets; further, the distribution of intersections is quite similar to that of sampling random error vectors and counting the intersections with $\mathcal{C}$, $\mathcal{N}$, and $2\mathcal{N}$. Our results indicate that these three sets are not sufficient in classifying vectors expected to cause decoding failures. Finally, we study the role of syndrome weight on the decoding behavior and conclude that the set of error vectors that lead to decoding failures differ from random vectors by having low syndrome weight.

Lightweight cryptography is characterized by the need for low implementation cost, while still providing sufficient security. This requires careful analysis of building blocks and their composition. SKINNY is an ISO/IEC standardized family of tweakable block ciphers and is used in the NIST lightweight cryptography standardization process finalist Romulus. We present non-trivial linear approximations of two- round SKINNY that have correlation one or minus one and that hold for a large fraction of all round tweakeys. Moreover, we show how these could have been avoided.

This note discusses lattice-based cryptography over the field with $p= 2^{64} - 2^{32} + 1$ elements, with an eye to supporting lattice-based cryptography operations in virtual machines such as Miden VM that operate natively over this field. It discusses how to support Dilithium and Falcon, two lattice-based signature scheme recently selected by the NIST PQC project; and proposes parameters for efficient public key encryption and publicly re-randomizable commitments modulo $p$.

The emergence of distributed digital currencies has raised the need for a reliable consensus mechanism. In proof-of-stake cryptocur- rencies, the participants periodically choose a closed set of validators, who can vote and append transactions to the blockchain. Each valida- tor can become a leader with the probability proportional to its stake. Keeping the leader private yet unique until it publishes a new block can significantly reduce the attack vector of an adversary and improve the throughput of the network. The problem of Single Secret Leader Election (SSLE) was first formally defined by Boneh et al. in 2020. In this work, we propose a novel framework for constructing SSLE proto- cols, which relies on secure multi-party computation (MPC) and satisfies the desired security properties. Our framework does not use any shuffle or sort operations and has a computational cost for N parties as low as O(N) of basic MPC operations per party. We improve the state-of-the- art for SSLE protocols that do not assume a trusted setup. Moreover, our SSLE scheme efficiently handles weighted elections. That is, for a total weight S of N parties, the associated costs are only increased by a factor of logS. When the MPC layer is instantiated with techniques based on Shamir’s secret-sharing, our SSLE has a communication cost of O(N2) which is spread over O(log N) rounds, can tolerate up to t < N/2 of faulty nodes without restarting the protocol, and its security relies on DDH in the random oracle model. When the MPC layer is instantiated with more efficient techniques based on garbled circuits, our SSLE re- quires all parties to participate, up to N − 1 of which can be malicious, and its security is based on the random oracle model.

Can we hope to provide provable security against model extraction attacks? As a step towards a theoretical study of this question, we unify and abstract a wide range of "observational" model extraction defense mechanisms -- roughly, those that attempt to detect model extraction using a statistical analysis conducted on the distribution over the adversary's queries. To accompany the abstract observational model extraction defense, which we call OMED for short, we define the notion of complete defenses -- the notion that benign clients can freely interact with the model -- and sound defenses -- the notion that adversarial clients are caught and prevented from reverse engineering the model. We then propose a system for obtaining provable security against model extraction by complete and sound OMEDs, using (average-case) hardness assumptions for PAC-learning. Our main result nullifies our proposal for provable security, by establishing a computational incompleteness theorem for the OMED: any efficient OMED for a machine learning model computable by a polynomial size decision tree that satisfies a basic form of completeness cannot satisfy soundness, unless the subexponential Learning Parity with Noise (LPN) assumption does not hold. To prove the incompleteness theorem, we introduce a class of model extraction attacks called natural Covert Learning attacks based on a connection to the Covert Learning model of Canetti and Karchmer (TCC '21), and show that such attacks circumvent any defense within our abstract mechanism in a black-box, nonadaptive way. Finally, we further expose the tension between Covert Learning and OMEDs by proving that Covert Learning algorithms require the nonexistence of provable security via efficient OMEDs. Therefore, we observe a "win-win" result by obtaining a characterization of the existence of provable security via efficient OMEDs by the nonexistence of natural Covert Learning algorithms.

We show that we can break SIDH in polynomial time, even with a random starting curve~$E_0$.

This work presents RPM, a scalable anonymous communication protocol suite using secure multiparty computation (MPC) with the offline-online model. We generate random, unknown permutation matrices in a secret-shared fashion and achieve improved (online) performance and the lightest communication and computation overhead for the clients compared to the state of art robust anonymous communication protocols. Using square-lattice shuffling, we make our protocol scale well as the number of clients increases. We provide three protocol variants, each targeting different input volumes and MPC frameworks/libraries. Besides, due to the modular design, our protocols can be easily generalized to support more MPC functionalities and security properties as they get developed. We also illustrate how to generalize our protocols to support two-way anonymous communication and secure sorting. We have implemented our protocols using the MP-SPDZ library suit and the benchmark illustrates that our protocols achieve unprecedented online phase performance with practical offline phases.

Multi-signatures are protocols that allow a group of signers to jointly produce a single signature on the same message. In recent years, a number of practical multi-signature schemes have been proposed in the discrete-log setting, such as MuSigT (CRYPTO'21) and DWMS (CRYPTO'21). The main technical challenge in constructing a multi-signature scheme is to achieve a set of several desirable properties, such as (1) security in the plain public-key (PPK) model, (2) concurrent security, (3) low online round complexity, and (4) key aggregation. However, previous lattice-based, post-quantum counterparts to Schnorr multi-signatures fail to satisfy these properties.

In this paper, we introduce MuSigL, a lattice-based multi-signature scheme simultaneously achieving these design goals for the first time. Unlike the recent, round-efficient proposal of Damgård et al. (PKC'21), which had to rely on lattice-based trapdoor commitments, we do not require any additional primitive in the protocol, while being able to prove security from the standard module-SIS and LWE assumptions. The resulting output signature of our scheme therefore looks closer to the usual Fiat--Shamir-with-abort signatures.

Secure multiparty computation can often utilize a trusted source of correlated randomness to achieve better efficiency. A recent line of work, initiated by Boyle et al. (CCS 2018, Crypto 2019), showed how useful forms of correlated randomness can be generated using a cheap, one-time interaction, followed by only "silent" local computation. This is achieved via a pseudorandom correlation generator (PCG), a deterministic function that stretches short correlated seeds into long instances of a target correlation. Previous works constructed concretely efficient PCGs for simple but useful correlations, including random oblivious transfer and vector-OLE, together with efficient protocols to distribute the PCG seed generation. Most of these constructions were based on variants of the Learning Parity with Noise (LPN) assumption. PCGs for other useful correlations had poor asymptotic and concrete efficiency.

In this work, we design a new class of efficient PCGs based on different flavors of the ring-LPN assumption. Our new PCGs can generate OLE correlations, authenticated multiplication triples, matrix product correlations, and other types of useful correlations over large fields. These PCGs are more efficient by orders of magnitude than the previous constructions and can be used to improve the preprocessing phase of many existing MPC protocols.

Impossible differential (ID) cryptanalysis is one of the most important attacks on block ciphers. The Mixed Integer Linear Programming (MILP) model is a popular method to determine whether a specific difference pair is an ID. Unfortunately, due to the huge search space (approximately $2^{2n}$ for a cipher with a block size $n$ bits), we cannot leverage this technique to exhaust all difference pairs, which is a well-known long-standing problem.

In this paper, we propose a systematic method to find all IDs for SPN block ciphers. The idea is to partition the whole difference pair space into lots of small disjoint sets, each of which has a representative difference pair. All difference pairs in one small set are possible if its representative pair is possible, and this can be conveniently checked by the MILP model. In this way, the overall search space is drastically reduced to a practical size by excluding the sets containing no IDs. We then examine the remaining difference pairs to identify all IDs (if some IDs exist). If our method cannot find any ID, the target cipher is proved free of ID distinguishers.

Our method works especially well for SPN ciphers with block size 64. We apply our method to SKINNY-64 and successfully find all 432 and 12 truncated IDs (we find all IDs but all of them can be assembled into certain truncated IDs) for 11 and 12 rounds, respectively. We also prove, for the first time, that 13-round SKINNY-64 is free of ID distinguishers even when considering the differential transitions through the Difference Distribution Table (DDT). Similarly, we find all 12 truncated IDs (all IDs are assembled into 12 truncated IDs) for 13-round CRAFT and prove there is no ID for 14 rounds. For SbPN cipher GIFT-64, we prove that there is no ID for 8 rounds.

For SPN ciphers with larger block sizes, we show that our idea is also useful to strengthen the current search methods. For example, if we consider the Sbox to be ideal and only consider the branch number information of the diffusion matrix, we can find all 6,750 truncated IDs for 6-round Rijndael-192 in 1 second and prove that there is no truncated ID for 7 rounds. Previously, we need to solve approximately $2^{48}$ MILP models to achieve the same goal. For GIFT-128, we exhausted all difference patterns that have an active superbox in the plaintext and ciphertext and proved there is no ID of such patterns for 8 rounds.

Although we have searched for a larger or even full space for IDs, no longer ID distinguishers have been found. This implies the reasonableness of the intuition that a small number (usually one or two) of active bits/words at the beginning and end of an ID will be the longest.

We give characterizations of IND\$-CPA security for a large, natural class of encryption schemes. Specifically, we consider encryption algorithms that invoke a block cipher and otherwise perform linear operations (e.g., XOR and multiplication by fixed field elements) on intermediate values. This class of algorithms corresponds to the Linicrypt model of Carmer & Rosulek (Crypto 2016). Our characterization for this class of encryption schemes is sound but not complete.

We then focus on a smaller subclass of block cipher modes, which iterate over the blocks of the plaintext, repeatedly applying the same Linicrypt program. For these Linicrypt block cipher modes, we are able to give a sound and complete characterization of IND\$-CPA security. Our characterization is linear-algebraic in nature and is easy to check for a candidate mode. Interestingly, we prove that a Linicrypt block cipher mode is secure if and only if it is secure against adversaries who choose all-zeroes plaintexts.

We obtain a black-box construction of non-interactive CCA commitments against non-uniform adversaries. This makes black-box use of an appropriate base commitment scheme for small tag spaces, variants of sub-exponential hinting PRG (Koppula and Waters, Crypto 2019) and variants of keyless sub-exponentially collision-resistant hash function with security against non-uniform adversaries (Bitansky, Kalai and Paneth, STOC 2018 and Bitansky and Lin, TCC 2018).

All prior works on non-interactive non-malleable or CCA commitments without setup first construct a "base" scheme for a relatively small identity/tag space, and then build a tag amplification compiler to obtain commitments for an exponential-sized space of identities. Prior black-box constructions either add multiple rounds of interaction (Goyal, Lee, Ostrovsky and Visconti, FOCS 2012) or only achieve security against uniform adversaries (Garg, Khurana, Lu and Waters, Eurocrypt 2021).

Our key technical contribution is a novel tag amplification compiler for CCA commitments that replaces the non-interactive proof of consistency required in prior work. Our construction satisfies the strongest known definition of non-malleability, i.e., CCA2 (chosen commitment attack) security. In addition to only making black-box use of the base scheme, our construction replaces sub-exponential NIWIs with sub-exponential hinting PRGs, which can be obtained based on assumptions such as (sub-exponential) CDH or LWE.

The Rank Decoding problem (RD) is at the core of rank-based cryptography. Cryptosystems such as ROLLO and RQC, which made it to the second round of the NIST Post-Quantum Standardization Process, as well as the Durandal signature scheme, rely on it or its variants. This problem can also be seen as a structured version of MinRank, which is ubiquitous in multivariate cryptography. Recently, [1,2] proposed attacks based on two new algebraic modelings, namely the MaxMinors modeling which is specific to RD and the Support-Minors modeling which applies to MinRank in general. Both improved significantly the complexity of algebraic attacks on these two problems. In the case of RD and contrarily to what was believed up to now, these new attacks were shown to be able to outperform combinatorial attacks and this even for very small field sizes.

However, we prove here that the analysis performed in [2] for one of these attacks which consists in mixing the MaxMinors modeling with the Support-Minors modeling to solve RD is too optimistic and leads to underestimate the overall complexity. This is done by exhibiting linear dependencies between these equations and by considering an Fqm version of these modelings which turns out to be instrumental for getting a better understanding of both systems. Moreover, by working over Fqm rather than over Fq, we are able to drastically reduce the number of variables in the system and we (i) still keep enough algebraic equations to be able to solve the system, (ii) are able to analyze rigorously the complexity of our approach. This new approach may improve the older MaxMinors approach on RD from [1,2] for certain parameters. We also introduce a new hybrid approach on the Support-Minors system whose impact is much more general since it applies to any MinRank problem. This technique improves significantly the complexity of the Support-Minors approach for small to moderate field sizes.

References:

[1] An Algebraic Attack on Rank Metric Code-Based Cryptosystems, Bardet, Briaud, Bros, Gaborit, Neiger, Ruatta, Tillich, EUROCRYPT 2020.

[2] Improvements of Algebraic Attacks for solving the Rank Decoding and MinRank problems, Bardet, Bros, Cabarcas, Gaborit, Perlner, Smith-Tone, Tillich, Verbel, ASIACRYPT 2020.

We study the problem of biometric-based authentication with template confidentiality. Typical schemes addressing this problem, such as Fuzzy Vaults (FV) and Fuzzy Extractors (FE), allow a server, aka Authenticator, to store “random looking” Helper Data (HD) instead of biometric templates in clear. HD hides information about the corresponding biometric while still enabling secure biometric-based authentication. Even though these schemes reduce the risk of storing biometric data, their correspondent authentication procedures typically require sending the HD (stored by the Authenticator) to a client who claims a given identity. The premise here is that only the identity owner - i.e., the person whose biometric was sampled to originally generate the HD - is able to provide the same biometric to reconstruct the proper cryptographic key from HD. As a side effect, the ability to freely retrieve HD, by simply claiming a given identity, allows invested adversaries to perform offline statistical attacks (a biometric analog for dictionary attacks on hashed passwords) or re-usability attacks (if the FE scheme is not reusable) on the HD to eventually recover the user’s biometric.

In this work we develop Oblivious Extractors: a new construction that allows an Authenticator to authenticate a user without requiring neither the user to send a biometric to the Authenticator, nor the server to send the HD to the client. Oblivious Extractors provide concrete security advantages for biometric-based authentication systems. From the perspective of secure storage, an oblivious extractor is as secure as its non-oblivious fuzzy extractor counterpart. In addition, it enhances security against aforementioned statistical and re-usability attacks. To demonstrate the construction’s practicality, we implement and evaluate a biometric-based authentication prototype using Oblivious Extractors.

The FIDO2 protocol is a globally used standard for passwordless authentication, building on an alliance between major players in the online authentication space. While already widely deployed, the standard is still under active development. Since version 2.1 of its CTAP sub-protocol, FIDO2 can potentially be instantiated with post-quantum secure primitives. We provide the first formal security analysis of FIDO2 with the CTAP 2.1 and WebAuthn 2 sub-protocols. Our security models build on work by Barbosa et al. for their analysis of FIDO2 with CTAP 2.0 and WebAuthn 1, which we extend in several ways. First, we provide a more fine-grained security model that allows us to prove more relevant protocol properties, such as guarantees about token binding agreement, the None attestation mode, and user verification. Second, we can prove post-quantum security for FIDO2 under certain conditions and minor protocol extensions. Finally, we show that for some threat models, the downgrade resilience of FIDO2 can be improved, and show how to achieve this with a simple modification.

Verifiable Data Streaming (VDS) enables a resource-limited client to continuously outsource data to an untrusted server in a sequential manner while supporting public integrity verification and efficient update. However, most existing VDS schemes require the client to generate all proofs in advance and store them at the server, which leads to a heavy computational burden on the client. In addition, all the previous VDS schemes can perform batch query (i.e., retrieving multiple data entries at once), but are subject to linear communication cost $l$, where $l$ is the number of queried data. In this paper, we first introduce a new cryptographic primitive named Double-trapdoor Chameleon Vector Commitment (DCVC), and then present an unbounded VDS scheme $\mathsf{VDS_1}$ with optimal communication cost in the random oracle model from aggregatable cross-commitment variant of DCVC. Furthermore, we propose, to our best knowledge, the first unbounded VDS scheme $\mathsf{VDS}_2$ with optimal communication and storage overhead in the standard model by integrating Double-trapdoor Chameleon Hash Function (DCH) and Key-Value Commitment (KVC). Both of our schemes enjoy constant-size public key. Finally, we demonstrate the efficiency of our two VDS schemes with a comprehensive performance evaluation.

The Massive Parallel Computing (MPC) model gained wide adoption over the last decade. By now, it is widely accepted as the right model for capturing the commonly used programming paradigms (such as MapReduce, Hadoop, and Spark) that utilize parallel computation power to manipulate and analyze huge amounts of data.

Motivated by the need to perform large-scale data analytics in a privacy-preserving manner, several recent works have presented generic compilers that transform algorithms in the MPC model into secure counterparts, while preserving various efficiency parameters of the original algorithms. The first paper, due to Chan et al. (ITCS ’20), focused on the honest majority setting. Later, Fernando et al. (TCC ’20) considered the dishonest majority setting. The latter work presented a compiler that transforms generic MPC algorithms into ones which are secure against semi-honest attackers that may control all but one of the parties involved. The security of their resulting algorithm relied on the existence of a PKI and also on rather strong cryptographic assumptions: indistinguishability obfuscation and the circular security of certain LWE-based encryption systems.

In this work, we focus on the dishonest majority setting, following Fernando et al. In this setting, the known compilers do not achieve the standard security notion called malicious security, where attackers can arbitrarily deviate from the prescribed protocol. In fact, we show that unless very strong setup assumptions as made (such as a programmable random oracle), it is provably impossible to withstand malicious attackers due to the stringent requirements on space and round complexity.

As our main contribution, we complement the above negative result by designing the first general compiler for malicious attackers in the dishonest majority setting. The resulting protocols withstand all-but-one corruptions. Our compiler relies on a simple PKI and a (programmable) random oracle, and is proven secure assuming LWE and SNARKs. Interestingly, even with such strong assumptions, it is rather non-trivial to obtain a secure protocol.

We present an attack on SIDH which does not require any endomorphism information on the starting curve. Our attack is not polynomial-time, but significantly reduces the security of SIDH and SIKE; our analysis and preliminary implementation suggests that our algorithm will be feasible for the Microsoft challenge parameters $p = 2^{110}3^{67}-1$ on a regular computer. Our attack applies to any isogeny-based cryptosystem that publishes the images of points under the secret isogeny, for example Séta [26] and B-SIDH [9]. It does not apply to CSIDH [8], CSI-FiSh [3], or SQISign [11].