Navigating organizational security policies is challenging due to their expansive and continuously evolving nature. LLMs can help streamline this process by understanding and interpreting complex policy language. This paper benchmarks this possibility and develops PARAG (Proactive Answering framework using Retrieval-Augmented Generation). It enhances policy comprehension by generating contextually accurate responses using a RAG-based approach. By structuring policy documents into semantically meaningful representations, the framework enables efficient indexing and retrieval of relevant information. This structured retrieval process allows the system to extract the most pertinent policy details, ensuring precise, policy-grounded answers without requiring extensive fine-tuning. To assess its effectiveness, we introduce an organization-agnostic dataset comprising 100 security policy-related question-answer pairs and evaluate the framework using policies from two real-world organizations. A comparative analysis of nine LLMs, including both black-box and open-source models, across 11 performance metrics demonstrates that GPT-4o-mini outperforms other black-box models, while Mistral-7B leads among open-source alternatives. These findings highlight the potential of RAG-based systems in transforming policy accessibility, enabling organizations to bridge the gap between policy documentation and practical compliance. Both the dataset and the framework code are hosted on the PARAG GitHub repository .