Sensitive information disclosure happens when an AI system shows someone information they should not see. That can be personal details about people, confidential details about an organization, passwords or other login details, or parts of the tool’s own hidden instructions, its system prompt. For example, a volunteer asks a chatbot a routine question and the answer includes another person’s home address.
The leaked information can come from three places: the data the model learned from (see training data extraction), the documents the tool is allowed to search, or what other users typed into it. The OWASP Top 10 for LLM Applications, a widely used list of AI security risks, includes this problem in its 2025 edition.
Why it matters
If your organization puts donor records, client files or staff details into an AI tool, those details could later show up in answers to someone else. Before you connect a tool to your files, ask the vendor who can see what you enter, whether it is kept or used for training, and how the tool keeps one user’s information out of another user’s answers.
Published Updated
Sources
- OWASP Top 10 for LLM Applications, checked October 27, 2026