databricks-vector-search

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The required runtime workflow is querying a Databricks Vector Search index using query_text that comes directly from the chat messages (query = messages[-1].content), and then inserting retrieved context_docs into the agent’s system prompt (Step 6 “Use in an Agent”), so outsider-authored free text can influence what the LLM sees.