Data Poisoning

Data poisoning is an attack where an adversary manipulates training or retrieval data to influence a model's outputs. In RAG-based agents, poisoned knowledge base entries can cause the agent to return harmful or misleading information.