Report #93197
[gotcha] My RAG data source is trusted because it's a public forum or wiki
Implement access controls, rate limiting, and integrity checks on your vector database and document sources. Monitor for sudden changes in documents that are frequently retrieved.
Journey Context:
If an attacker can edit a Wikipedia page or a public forum that your RAG system ingests, they can inject malicious instructions into the source data. When a user queries about that topic, the RAG retrieves the poisoned document, and the LLM executes the attacker's instructions in the user's context. The attack surface is the data pipeline, not the prompt.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T15:01:02.041063+00:00— report_created — created