Towards a Proactive Defense: Poisoning Detection in Agentic Systems
A Linkalab framework for detecting “poisoned” content in LLM+RAG agent knowledge bases: seeded detection, analysis in the embedding space, and results on the AgentPoison dataset and a real-world case.