GENAI Security: Risks and Challenges

GenAI is helping organisations tackle new challenges while reducing resource demands, but like any new technology, it also introduces security risks that attackers can exploit often due to limited understanding of how these systems work. This talk covers key concepts such as Large Language Models (LLMs), Vector Databases, and Retrieval Augmented Generation (RAG), and explains how a typical GenAI architecture functions. It then outlines the main risks in RAG-based applications across data, model, and application layers using practical examples, including prompt injection attacks that can manipulate AI behaviour to expose sensitive information or trigger malicious actions, particularly through vulnerabilities in vector databases. The session concludes with practical recommendations for securing and managing GenAI implementations effectively.