For Heads-of · Practitioner

Hidden context exposure

Content the operator assumed was hidden from the user, such as retrieval context, RAG schemas, or policy logic, is exposed or inferred through the model, beyond just the system prompt.

  • medium
  • llm
  • rag
  • context
  • owasp-llm

How it happens

A RAG pipeline's retrieval context, schema, or the policy logic steering a model's behaviour is assumed to be invisible to the end user, but the model can be induced to reveal or reconstruct it through the same probing techniques that extract a system prompt.

Why it matters

Broader than a leaked system prompt: it can expose the shape of an internal knowledge base, the rules an agent is following, or business logic the operator never intended to be user-facing, handing an attacker a map of what to target next.

Mitigating controls

The controls that address this risk, ranked by effectiveness.

Framework and clause references

FrameworkClauseTitle
OWASP Top 10 for LLM ApplicationsLLM08Hidden Context Exposure