Skip to content
AI Defense Lab

RAG Workbench

Ingest documents, query with evidence provenance, and scan for RAG poisoning attacks.

RAG (Retrieval-Augmented Generation) lets AI systems look up information from a knowledge base before answering. If an attacker poisons that knowledge base, the AI will give wrong or dangerous answers.
Loading sources...

Challenge: Can You Poison the Knowledge Base?

Try crafting a document that poisons the knowledge base. Ingest a document with contradicting claims or hidden instructions, then query to see if it appears in results. Can you get a poisoned chunk to rank highly? Check the Scan tab afterward to see if the poisoning detector catches your attempt.