Skip to content
Open access

NASC: Neuro-symbolic Architecture for Smart Contracts Vulnerability Detection

Sep 2026 · Brazilian Symposium on Systematic and Automated Software Testing · 0 citations · 18 references

Abstract

Smart contracts control digital assets under conditions of immutability in blockchain networks, making vulnerability detection an important task in decentralized systems security. Traditional static analysis tools generally rely on specific syntactic patterns, which can make it difficult for developers to interpret the results. This work presents NASC, a Neuro-symbolic agent architecture for vulnerability analysis in Solidity smart contracts. The approach combines a language model-based planner agent, a symbolic core responsible for converting the source code into first-order logic facts, and an LLM agent orchestration module based on the BMAD (Build More Architect Dreams) method that enables the agent to read, write, and run the symbolic core of NASC. The detection mechanism is performed by declarative rules implemented in Prolog, while an analyst agent transforms the results into structured natural language reports. The architecture was evaluated using three literature datasets. Our results show that NASC is capable of identifying different vulnerability patterns and producing interpretable reports, achieving over 80% of recall in 2 of 3 datasets analyzed. Furthermore, the study highlights the potential of integrating logical inference and language model-based agents to support security analysis processes.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.