Sensitive Information Disclosure

Large Language Models generate responses based on patterns learned during training and on data retrieved at runtime. Unlike traditional databases, they do not enforce strict row-level access rules by default. This creates a new class of confidentiality risk: sensitive information disclosure. In LLM systems, disclosure does not usually happen because a model "decides" to reveal a secret. It happens because sensitive data was allowed to enter the system's context window.

In this room, you will explore how data leaks through the retrieval and logging layers of AI systems. You will start by examining how sensitive information disclosure is categorised under OWASP LLM02 and why it differs from poisoning and prompt injection. From there, you will investigate common disclosure scenarios, embedding-level risks, retrieval pipeline
failures, and access control strategies. You will then apply defensive safeguards before testing everything hands-on in a practical lab exercise against a simulated RAG system.

These exposures are categorised under OWASP LLM02 – Sensitive Information Disclosure. They focus on protecting private data, proprietary logic, and confidential documents from being exposed through AI outputs.

Learning Objectives

By completing this task, you will be able to:

• Define sensitive information disclosure (LLM02)
• Distinguish between parametric memory and retrieval-based leakage
• Identify architectural points where data can leak
• Understand why LLM confidentiality is a system design issue
• Prepare to analyse real disclosure scenarios in the next task

Prerequisites

This room is part of a broader AI Security path. It is recommended that you complete this room in the intended order to establish core fundamentals. At a minimum, you should be familiar with the concepts covered in the RAG Security Fundamentals and Data Poisoning in RAG Systems rooms. You should also be familiar with:

• How LLMs generate responses
• High-level knowledge of vector databases (recommended)

No machine learning or mathematical background is required.