Vinu Sankar Sadasivan, a PhD candidate at the University of Maryland and Student Researcher at Google DeepMind, dives into the crucial themes of AI robustness and security. He discusses the challenges of jailbreaking multimodal models and explores innovative watermarking techniques for identifying AI-generated content. Vinu highlights the complexities of red teaming practices and automated vulnerability exploitation, showcasing the ongoing battle between AI manipulators and defenders. This engaging session sheds light on the future of safe AI applications across various fields.