Vulnerability discovery in C/C++ typically forces a trade-off. Traditional static analysers are heavyweight and need a working build before they can run. LLMs can read source code directly but are prone to inventing details.
NeuroLog is a neuro-symbolic framework that tries to combine the strengths of both. It uses LLMs to extract facts from source code, then applies Datalog rules and SMT solvers to reason over those facts. This yields a decision process that can be audited, rather than a black-box answer.
Because the approach is grounded in formal reasoning, it aims to give analysts a way to verify how a vulnerability was found. The paper positions this as a step beyond free-form LLM output, while avoiding the build requirement of traditional analysers.