Security Assurance Based on Large Language Models

Dependable systems require continuous assurance across distributed infrastructures to validate their non-functional properties. Yet selecting and configuring assurance probes remains complex, demanding expert knowledge of standards and mechanisms, and is difficult to scale across the cloud continuum. This paper proposes a methodology that leverages lightweight Large Language Models (LLMs) and a Retrieval-Augmented Generation (RAG) approach to help operators identify and configure suitable assurance probes. We introduce an architecture that enriches operator queries with evidence and compliance context before processing them through an LLM-based assistant. Evaluation, including an LLM-as-a-Judge analysis, shows that this approach reduces operator effort while preserving oversight, strengthening assurance processes in dynamic, distributed environments.