Patrick Walsh talks about interesting hidden risks in LLM’s and RAG workflows. He demonstrates how to extract sensitive data, such as personally identifiable information and social security numbers these systems were trained on, with real-world attacks.