Paper 2025/2328

SNARGs for NP via Fiat--Shamir in the Plain Model

Abstract

We consider constructions of succinct non-interactive arguments (SNARGs) for NP in the standard model. Specifically, we revisit the seminal Micali transformation (applying Fiat-Shamir to Kilian's protocol), which has traditionally only been analyzed in the random oracle model. We show that the Micali framework can be successfully instantiated in the standard model by leveraging a new interaction between two primitives: a PCP satisfying a property we term shadow soundness, and a vector commitment scheme satisfying function statistical binding. We prove a general theorem stating that any language admitting a suitable shadow PCP combined with a compatible vector commitment yields a secure SNARG. We instantiate this paradigm using sub-exponential indistinguishability obfuscation (iO) and sub-exponential learning with error (LWE) to obtain a SNARG for all of NP. Our result serves as the first concrete validation of the Micali blueprint, and in particular of the Fiat-Shamir transformation, in the standard model. As a corollary, we refute "universal" attacks on the Micali framework by demonstrating that there exist concrete instantiations of the underlying components for which the transformation is sound.