It seems to be a generally acknowledged fact that you should never trust a computer and that you should trust the person operating the computer even less. This in particular becomes a problem when the party that you do not trust is one which is separated from you and is one on which you depend, e.g. because that party is the holder of some data or some capability that you need in order to operate correctly. How do you perform a given task involving interaction with other parties while allowing these parties a minimal influence on you and, if privacy is an issue, allowing them to learn as little about you as possible. This is the general problem of secure multiparty computation. The usual way of formalizing the problem is to say that a number of parties who do not trust each other wish to compute some function of their local inputs, while keeping their inputs as secret as possible and guaranteeing the correctness of the output. Both goals should be obtained even if some parties stop participating or some malevolent coalition of the parties start deviating arbitrarily from the agreed protocol. The task is further complicated by the fact that besides their mutual distrust, nor do the parties trust the channels by which they communicate. A general solution to the secure multiparty computation problem is a compiler which given any feasible function describes an efficient protocol which allows the parties to compute the function securely on their local inputs over an open network. Over the past twenty years the secure multiparty computation problem has been the subject of a large body of research, both research into the models of multiparty computation and research aimed at realizing general secure multiparty computation. The main approach to realizing secure multiparty computation has been based on verifiable secret sharing as computation, where each party holds a secret share of each input and during the execution computes secret shares of all intermediate values. This approach allows the parties to keep all inputs and intermediate values secret and to pool the shares of the output values to learn exactly these values. More recently an approach based on joint homomorphic encryption was introduced, allowing for an efficient realization of general multiparty computation secure against an eavesdropper. A joint encryption scheme is an encryption scheme where all parties can encrypt, but where it takes the participation of all parties to open an encryption. The computation then starts by all parties broadcasting encryptions of their inputs and progresses through computing encryptions of the intermediary values using the homomorphic properties of the encryption scheme. The encryptions of the outputs can then be jointly decrypted. In this dissertation we extend this approach by using threshold homomorphic encryption to provide full-fledged security against an active attacker which might control some of the parties and which might take over parties during the course of the protocol execution. As opposed to a joint encryption scheme a threshold encryption scheme only requires that a given number of the parties are operating correctly for decryption to be possible. In this dissertation we show that threshold homomorphic encryption allows to solve the secure general multiparty computation problem more efficiently than previous approaches to the problem. Starting from an open point-to-point network there is a long way to general secure multiparty computation. The dissertation contains contributions at several points along the way. In particular we investigate how to realize secure channels. We also show how threshold signature schemes can be used to efficiently realize a broadcast channel and how threshold cryptography can be used to provide the parties of the network with a source of shared randomness. All three tools are well-known to be extremely powerful resources in a network, and our principle contribution is to provide efficient realizations. The dissertation also contains contributions to the definitional part of the field of multiparty computation. Most significantly we augment the recent model of universally composable security with a formal notion of what it means for a protocol to only realize a given problem under a given restricted input-output behavior of the environment. This e.g. allows us to give the first specification of a universally composable zero-knowledge proof of membership which does not at the same time require the proof system to be a proof of knowledge.
It seems to be a generally acknowledged fact that you should never trust a computer and that you should trust the person operating the computer even less. This in particular becomes a problem when the party that you do not trust is one which is separated from you and is one on which you depend, e.g. because that party is the holder of some data or some capability that you need in order to operate correctly. How do you perform a given task involving interaction with other parties while allowing these parties a minimal influence on you and, if privacy is an issue, allowing them to learn as little about you as possible. This is the general problem of secure multiparty computation. The usual way of formalizing the problem is to say that a number of parties who do not trust each other wish to compute some function of their local inputs, while keeping their inputs as secret as possible and guaranteeing the correctness of the output. Both goals should be obtained even if some parties stop participating or some malevolent coalition of the parties start deviating arbitrarily from the agreed protocol. The task is further complicated by the fact that besides their mutual distrust, nor do the parties trust the channels by which they communicate. A general solution to the secure multiparty computation problem is a compiler which given any feasible function describes an efficient protocol which allows the parties to compute the function securely on their local inputs over an open network. Over the past twenty years the secure multiparty computation problem has been the subject of a large body of research, both research into the models of multiparty computation and research aimed at realizing general secure multiparty computation. The main approach to realizing secure multiparty computation has been based on verifiable secret sharing as computation, where each party holds a secret share of each input and during the execution computes secret shares of all intermediate values. This approach allows the parties to keep all inputs and intermediate values secret and to pool the shares of the output values to learn exactly these values. More recently an approach based on joint homomorphic encryption was introduced, allowing for an efficient realization of general multiparty computation secure against an eavesdropper. A joint encryption scheme is an encryption scheme where all parties can encrypt, but where it takes the participation of all parties to open an encryption. The computation then starts by all parties broadcasting encryptions of their inputs and progresses through computing encryptions of the intermediary values using the homomorphic properties of the encryption scheme. The encryptions of the outputs can then be jointly decrypted. In this dissertation we extend this approach by using threshold homomorphic encryption to provide full-fledged security against an active attacker which might control some of the parties and which might take over parties during the course of the protocol execution. As opposed to a joint encryption scheme a threshold encryption scheme only requires that a given number of the parties are operating correctly for decryption to be possible. In this dissertation we show that threshold homomorphic encryption allows to solve the secure general multiparty computation problem more efficiently than previous approaches to the problem. Starting from an open point-to-point network there is a long way to general secure multiparty computation. The dissertation contains contributions at several points along the way. In particular we investigate how to realize secure channels. We also show how threshold signature schemes can be used to efficiently realize a broadcast channel and how threshold cryptography can be used to provide the parties of the network with a source of shared randomness. All three tools are well-known to be extremely powerful resources in a network, and our principle contribution is to provide efficient realizations. The dissertation also contains contributions to the definitional part of the field of multiparty computation. Most significantly we augment the recent model of universally composable security with a formal notion of what it means for a protocol to only realize a given problem under a given restricted input-output behavior of the environment. This e.g. allows us to give the first specification of a universally composable zero-knowledge proof of membership which does not at the same time require the proof system to be a proof of knowledge.
The notion of secure computation is central to cryptography. Introduced in the seminal works of Yao [FOCS'82, FOCS'86] and Goldreich, Micali and Wigderson [STOC'87], secure multiparty computation allows a group of (mutually) distrustful parties to jointly compute any functionality over their individual private inputs in such a manner that the honest parties obtain the correct outputs and no group of malicious parties learn anything beyond their inputs and the prescribed outputs. General feasibility results for secure computation were given by Yao [FOCS'86] and Goldreich et al. [STOC'87] more than two decades ago. Subsequent to these works, designing secure computation protocols that can tolerate more powerful adversaries and satisfy stronger notions of security has been a very active area of research. In this dissertation, we study two such new frontiers in the area of secure computation.
In the first part of this dissertation, we initiate a study of designing leakage-resilient interactive protocols. Specifically, we consider the scenario where an adversary, in addition to corrupting a subset of parties, can leak (potentially via physical attacks) arbitrary information from the secret state of any honest party. This is in contrast to the standard notion of secure computation where it is assumed that the adversary only has black-box'' access to the honest parties. In particular, we formalize a meaningful definition of leakage-resilient zero knowledge proof systems and provide constructions that satisfy our notion. We also discuss various applications of our results.
The second part of this dissertation concerns with the general question of whether it is possible to securely run cryptographic protocols over an insecure network environment such as the Internet. It is well-known that the standard notion of secure computation is only relevant to the stand-alone setting where a single protocol is being executed in isolation; as such it does not guarantee security when multiple protocol sessions may be executed concurrently under the control of an adversary who is present across all sessions. We consider the open problem of constructing secure password-based authenticated key exchange protocols in such a setting in the plain model (i.e., without assuming any trusted infrastructure or random oracles). We give the first construction of such a protocol based on standard cryptographic assumptions. Our results are in fact much more general, and extend to other functionalities w.r.t. a (necessarily) weakened notion of concurrently secure computation.
The results presented in this dissertation stem from two papers which are joint works with Sanjam Garg and Amit Sahai, and with Vipul Goyal and Rafail Ostrovsky, respectively.
Secure multi-party computation is a conceptual framework in which distrusting parties engage in a protocol to securely perform a computational task. Depending on the precise model of security, different sets of tasks admit secure protocols. We take a complexity-theoretic approach to studying the inherent difficulty of securely realizing tasks in various standard security models. * We give the first alternate characterization of secure realizability in the framework of universally composable (UC) security. This is the first characterization in any model to consider completely arbitrary computational tasks and completely arbitrary communication channels. * The most long-standing class of computational tasks studied are those in which two parties evaluate a deterministic function. For these tasks, we give the first complete, combinatorial characterizations of secure realizability in the passive, standalone, and universally composable security models, against computationally unbounded adversaries. * Say that a task G has ``as much cryptographic complexity'' as another task F if there is a secure protocol for F that uses access to a secure implementation of G. We show that there is an infinite hierarchy of tasks with strictly increasing cryptographic complexities, with respect to computationally unbounded security. We also show that there exist tasks whose cryptographic complexities are incomparable. * In contrast, we show that under a standard cryptographic assumption, there exist only two distinct levels of cryptographic complexity with respect to polynomial-time security. Every task either has a trivial protocol using plain communication channels, or is complete (i.e., given access to a secure implementation of this task, there are secure protocols for all other tasks). This is the first result to derive a characterization of completeness for a class of arbitrary interactive tasks. In light of these characterizations, the only tasks which are securely realizable in the demanding framework of universal composition are those related to secure communication. Indeed, the framework has been used to define the security of encryption schemes, which has allowed for modular design and analysis of protocols. We consider a similar approach for homomorphic encryption schemes. A homomorphic scheme is one in which anyone can obtain an encryption of f(m_1, ..., m_n), given only the encryptions of unknown messages m_1, ..., m_n, for a specific set of functions f. * We give a construction of a homomorphic encryption scheme in which the allowed homomorphic operation is as full-featured as possible --- namely, one can derive a correctly-distributed encryption of f(m) given an encryption of unknown message m, for some functions f --- yet it is computationally infeasible to generate a ciphertext that is related to other ciphertexts in any other way. Our contributions involve developing new appropriate security definitions as well as new constructions. * We show that schemes with such powerful security guarantees can be used to build conceptually simple, efficient, UC-secure protocols for verifiable computations on encrypted data. We show protocols for two tasks related to aggregating encrypted data.
Introduction In Chapter 1 we gave an introduction to the multiparty computation problem and explained what it intuitively means for a protocol to compute a given function securely, namely, we want a protocol that is correct and private. However, we only considered three players, we only argued that a single player does not learn more than he or she is supposed to, and finally, we assumed that all players would follow the protocol. In this chapter we will consider a more general solution to multiparty computation, where we remove the first two restrictions: we will consider any number n ≥ 3 of players, and we will be able to show that as long as at most t < n /2 of the players go together after the protocol is executed and pool all their information, they will learn nothing more than their own inputs and the outputs they were supposed to receive, even if their computing power is unbounded. We will still assume, however, that all players follow the protocol. This is known as semihonest or passive security . To argue security in this chapter, we will use a somewhat weak but very simple definition that only makes sense for semihonest security. We then extend this in Chapter 4 to a fully general model of what protocols are and what security means. Throughout this chapter we will assume that each pair of players can communicate using a perfectly secure channel, so if two players exchange data, a third player has no information at all about what is sent. Such channels might be available because of physical circumstances, or we can implement them relative to a computational assumption using cryptography. For more details on this, see Chapter 7. We end the chapter by showing that the bound t < n /2 is optimal: if t ≥ n /2, then some functions cannot be computed securely in the model mentioned, that is, obtaining perfect security by using only secure point-to-point communication as the communication resource. Secret Sharing Our main tool to build the protocol will be secret-sharing schemes. The theory of secret-sharing schemes is a large and interesting field in its own right with many applications to multiparty computation (MPC), and we look at this in more detail in Chapter 11, where a formal definition of the notion as well as a general treatment of the theory of secret sharing can be found.
The aim of this paper is to present some of the recent progress in efficient secure multiparty computation (MPC). In MPC we have a set of parties owning a set of private inputs. The par ties want to compute a function of their inputs, but they do not trust each other, therefore they need a cryptographic protocol to perform the computation in a way that 1) the output is correct and 2) cheating parties will not be able to learn any information about the honest parties inputs. Even though this problem has been formulated and essentially solved almost 30 years ago, practical solutions that can be relevant for real-world applications have been discovered only in the last few years. We will present some of these advances, trying to explain to a non-specialized audience the significance of the several existing security notions.
Consider several parties that do not trust each other, yet they wish to correctly compute some common function of their local inputs while keeping these inputs private. This problem is known as "Secure Multi-Party Computation", and was introduced by Andrew Yao in 1982. Secure multi-party computations have some real world examples like electronic auctions, electronic voting or fingerprinting. In this thesis we consider the case where there are only two parties involved. This is known as "Secure Two-Party Computation". If there is a trusted third party called Carol, then the problem is pretty straightforward. The participating parties could hand their inputs in Carol who can compute the common function correctly and could return the outputs to the corresponding parties. The goal is to achieve (almost) the same result when there is no trusted third party. Cryptographic protocols are designed in order to solve these kinds of problems. These protocols are analyzed within an appropriate model in which the behavior of parties is structured. The basic level is called the Semi-Honest Model where parties are assumed to follow the protocol specification, but later can derive additional information based on the messages which have been received so far. A more realistic model is the so-called Malicious Model. The common approach is to first analyze a protocol in the semi-honest model and then later extend it into the malicious model. Any cryptographic protocol for secure two-party computation must satisfy the following security requirements: correctness, privacy and fairness. It must guarantee the correctness of the result while preserving the privacy of the parties’ inputs, even if one of the parties is malicious and behaves arbitrarily throughout the protocol. It must also guarantee fairness. This roughly means that whenever a party aborts the protocol prematurely, he or she should not have any advantage over the other party in discovering the output. The main question for researchers is to construct new protocols that achieve the above mentioned goals for secure multi-party computation. Of course, such protocols must be secure in a given model, as well as be as efficient as possible. In 1986, Yao presented the first general protocol for secure two-party computation which was applicable only to the semi-honest model. He uses a tool called "Garbled Circuit". Yao’s protocol uses the underlying primitives ("Pseudorandom Generator" and "Oblivious Transfer") as blackboxes which lead to efficient results. After Yao’s work many variants and improvements have been proposed for the malicious model. In this thesis, we design several new protocols for secure two-party computation based on Yao’s garbled circuit. Before we present the details of our new designs, we first show several weaknesses, security flaws or problems with the existing protocols in the literature. We first work in the semi-honest model and then extend it into the malicious model by presenting new protocols. Finally we add fairness to our protocol. Oblivious transfer (OT) is a fundamental primitive in modern cryptography which is useful for implementing protocols for secure multi-party computation. We study several variants of oblivious transfer in this thesis. We present a new protocol for the so-called "Committed OT". This protocol is very efficient in the sense that it is quite good in comparison to the most efficient committed OT protocols in the literature. The abovementioned flaw with the use of OT can be fixed with our committed oblivious transfer protocol. Furthermore, it is more general than all previous protocols, and, therefore, it is of independent interest. We also deal with fairness in this thesis. For protocols based on garbled circuit, so far only Benny Pinkas has presented a protocol in the literature for achieving fairness. We show a subtle problem with this protocol where the privacy of the inputs of one party can be compromised. We also describe this problem in detail which is in fact related to the fairness, and finally propose a more efficient scheme that does achieve fairness.
Secure Multiparty Computation (MPC) is an expression encompassing cryptographic techniques that allow mutually distrustful parties to jointly perform computations over private data, ensuring correctness and preserving privacy despite malicious behavior. Guaranteed output is the strongest security goal for MPC: protocols with this property always deliver results. This is in contrast with weaker notions that allow malicious aborts. The seminal works of Ben-Or, Goldwasser, and Wigderson (STOC’88) and Rabin and Ben-Or (STOC’89) showed interactive protocols to compute any function with guaranteed output and perfect security if fewer than one third of parties are corrupted, or with statistical security (with a small error probability) if fewer than half are corrupted. These results hold when the set of participants is fixed throughout the execution. Ostrovsky and Yung’s proactive model (PODC’91) relaxes this assumption by allowing the set of corrupted parties to evolve over time, better capturing the dynamics of long-lived computations, where machines may be periodically compromised and recovered. The recent YOSO (You Only Speak Once) and fluid MPC models (CRYPTO’21) push this ambitious modelling even further by allowing for fully dynamic participation, with parties joining and leaving the computation at will. In this setting, known protocols with guaranteed output fall short of optimal resilience or rely on non-standard communication and adversarial assumptions. Matching the classical BGW and Rabin–Ben-Or bounds in these settings remained an open problem. We resolve these questions within the Layered MPC abstraction of David et al. (CRYPTO’23), in which the interaction pattern of a protocol is restricted to be a layered graph, meaning parties are partitioned into a sequence of disjoint sets (layers) and can send private and broadcast messages to parties in immediately subsequent layers. We match the BGW and Rabin and Ben-Or bounds in this model, fully characterizing the feasibility landscape of Layered MPC with guaranteed output. As a stepping stone towards our statistically secure protocol, we construct a more efficient and conceptually simpler computationally secure protocol relying on linear non-interactive commitments. Our results yield optimally-resilient and maximally-proactive MPC protocols with guaranteed output for both perfect and statistical security.