Paper 2025/995

NIZK Amplification via Leakage-Resilient Secure Computation

Abstract

Suppose that we are given a weak \emph{Non-Interactive Zero-Knowledge} (NIZK) proof system for NP with non-negligible soundness and zero-knowledge errors, denoted by $\alpha$ and $\beta$, respectively. Is it possible to to reduce these errors to a negligible level? This problem, known as NIZK amplification, was introduced by Goyal, Jain, and Sahai (Crypto'19) and was further studied by Bitansky and Geier (Crypto'24). The latter work provides amplification theorems for proofs and arguments, assuming the existence of one-way functions and public-key encryption, respectively. Unfortunately, their results only apply when the security level, $1 - (\alpha + \beta)$, is a constant bounded away from zero. Amplifying NIZK with an inverse polynomial security level remains an open problem and was stated as the main open question in both works. In this work, we resolve the NIZK amplification problem and show how to amplify any non-trivial NIZK proof system that has a noticeable, inverse-polynomial level of security. As in previous works, we amplify proofs and arguments assuming the existence of one-way functions and public-key encryption, respectively. Furthermore, assuming the existence of collision-resistant hash functions, we preserve, for the first time, properties such as statistical zero-knowledge and proof succinctness. Our main technical contribution is a new \emph{leakage-resilient secure multiparty} protocol that computes any public-output functionality with information-theoretic security against an adversary that corrupts an arbitrary subset of parties and obtains bounded leakage from each honest party. Our protocol operates in the pairwise correlated randomness model. Previous works relied on stronger setup assumptions in the form of $n$-wise correlations and either supported a smaller corruption threshold or suffered from an exponential dependency on the number of parties. To transform our protocol into a NIZK amplifier, we introduce a new intermediate notion of \emph{leakage-resilient NP secret sharing}, that may be of independent interest.