FlagThis

Abstract

This presentation provides a comprehensive overview of the security landscape surrounding Generative AI (GenAI). It explores the fundamental architecture of deep learning and neural networks to explain how vulnerabilities emerge within the model, pipeline, and output layers. The talk details critical threats including indirect prompt injection, model supply chain poisoning, and deepfake-driven social engineering. Key takeaways include actionable defense frameworks such as Retrieval Augmented Generation (RAG), human-in-the-loop protocols, and robust data governance to protect against PII leakage and operational instability.