sensitive information disclosure
An application revealing personal data, secrets or confidential business information through a model's output.
Leaked personal data or secrets cause legal, financial and reputational harm that cannot be undone.
In more detail
It is second in the OWASP Top 10 for LLM Applications 2026. Keep secrets out of prompts, filter what retrieval may return by the user's permissions, redact personal data on the way in and out, and assume anything in the context can be repeated.
Related terms
- PII redaction — Finding personal data such as names, emails, card numbers and phone numbers in text and replacing it before it is stored, logged or sent to a model.
- hidden context exposure — The OWASP 2026 entry, formerly system prompt leakage, warning that system prompts and other hidden context can be extracted by users or attackers.
- data exfiltration — Secretly copying data out of a system to somewhere an attacker controls.
Taught in
- Threat-Model An LLM App — AI Engineering: LLM Apps, RAG & Agents
References
- OWASP GenAI Security Project: Top 10 for LLM Applications 2026
- Microsoft Presidio: data protection and de-identification SDK