Paper 2023/415

Maximally-Fluid MPC with Guaranteed Output Delivery

Abstract

To overcome the limitations of traditional secure multi-party computation (MPC) protocols that consider a static set of participants, in a recent work, Choudhuri et al. [CRYPTO 2021] introduced a new model called Fluid MPC, which supports {\em dynamic} participants. Protocols in this model allow parties to join and leave the computation as they wish. Unfortunately, known fluid MPC protocols (even with strong honest-majority), either only achieve security with abort, or require strong computational and trusted setup assumptions. In this work, we also consider the "hardest" setting --- called the maximally-fluid model --- where each party can leave the computation after participating in a single round. We study the problem of designing maximally-fluid MPC protocols that achieve security with {guaranteed output delivery}, and obtain the following main results: 1. We design a perfectly secure maximally-fluid MPC protocol, that achieves guaranteed output delivery against unbounded adversaries who are allowed to corrupt less than a third of the parties in every round/committee. 2. For the case where the adversary is allowed to corrupt up to half of the parties in each committee, we present a new computationally secure maximally-fluid MPC protocol with guaranteed output delivery. Unlike prior works that require correlated setup and NIZKs, our construction only uses a common random string setup and is based on linearly-homomorphic equivocal commitments.

Note: The impossibility result in the previous version of our paper was incorrect. This is a revised version that does not include that result.