cloudflare-workers-ai

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

• Third-party content exposure detected (medium risk: 0.65). In the RAG template (templates/ai-embeddings-rag.ts), the runtime builds context from match.metadata?.text returned by env.VECTORIZE.query(...) and injects it into the LLM prompt as system content (content: ...\n\nContext:\n${context}), so any outsider-authored text previously stored in Vectorize (e.g., from documents the user didn’t author) can flow into the model context.