Back to OWASP Top 10#2 High Risk

LLM02:2025 Sensitive Information Disclosure

LLMs risk exposing sensitive data, proprietary algorithms, or confidential details through their output, resulting in unauthorized data access, privacy violations, and intellectual property breaches.

Vulnerability Overview

Sensitive information can affect both the LLM and its application context, including PII, financial details, health records, confidential business data, security credentials, and legal documents.

Impact Level

High

Data at Risk

PII, Business Data, Algorithms

Attack Vector

Model Output

Common Vulnerability Examples

1. PII Leakage

Personal identifiable information (PII) may be disclosed during interactions with the LLM, exposing names, addresses, social security numbers, or other sensitive personal data.

Example: User asks for help with a document, and the LLM accidentally reveals another user's personal information that was in its training data.

2. Proprietary Algorithm Exposure

Poorly configured model outputs can reveal proprietary algorithms or data. This can expose models to inversion attacks where attackers extract sensitive information.

Example: The 'Proof Pudding' attack (CVE-2019-20634) demonstrated how disclosed training data facilitated model extraction and inversion attacks.

3. Sensitive Business Data Disclosure

Generated responses might inadvertently include confidential business information, trade secrets, or internal processes.

Example: A customer service chatbot accidentally reveals internal pricing strategies or unreleased product information.