What you will practice
- Identify poisoned retrieval content
- Separate data from instructions
- Apply provenance and grounding checks
Learning outcomes
- Trace an indirect prompt injection path
- Explain why retrieved text is untrusted
- Recommend secure retrieval controls
Prerequisites
Basic familiarity with retrieval-augmented generation is helpful but not required.
Read the RAG Security lesson firstLaunch the cloud playground
Sign in to start an isolated cloud lab. Each included lab has three sessions: five minutes on Free or fifteen minutes on Launch.
Need the concepts again? Review What Is RAG Security? Attacks, Poisoning, and Defenses.
