CyberGuard dictionary / AI Security
Data Prompt Leakage
Data prompt leakage is unintended disclosure of sensitive material supplied to, retrieved by, or generated within an AI application prompt context.
In simple terms
A prompt may contain customer records, internal instructions, or confidential documents. Those details can escape if access checks or output controls fail.
Technical explanation
Leakage paths include cross-tenant retrieval, overbroad tool access, logging, and model responses that repeat protected context. Data minimization, authorization at retrieval time, and output review reduce risk.
Protected context can escape through responses, tool calls, logs, or retrieval across users. The application should minimize what the model receives and enforce authorization before data enters the prompt.
Examples in practice
A shared assistant includes part of a private contract in an answer to a user who only asked a general pricing question.
A support assistant is asked to summarize a public policy but includes a private contract excerpt. The team traces the retrieval filter error and checks logs for other affected requests.
Why it matters
Treat prompt context as sensitive application data with its own retention and access rules.
VERIFY & LEARN MORE
Sources and further reading
- genai.owasp.orghttps://genai.owasp.org/llm-top-10/↗
- genai.owasp.orghttps://genai.owasp.org/llmrisk/llm082025-vector-and-embedding-weaknesses/↗
Published 2026-10-15 · Last updated 2026-10-15 · CyberGuard editorial content. Source links provide context; they do not imply endorsement.