This article explores a novel approach to dynamically score the confidence of AI‑generated responses to security questionnaires, leveraging real‑time evidence feedback, knowledge graphs, and LLM orchestration to improve accuracy and auditability.