[AIP AI Security and Privacy team seminar] Towards Safe AI: : Cybersecurity Tools and Techniques for AI. Luis Ibanez-Lissen (Universidad Carlos III de Madrid))
Title: Towards Safe AI: Luis Ibanez-Lissen (Universidad Carlos III de Madrid))
Abstract: In this presentation, I will discuss key research contributions, focusing on cybersecurity techniques for AI. I will cover a range of approaches related to fake news detection and LLM red teaming, highlighting how these methods can enhance AI safety. Finally, I will present my latest work on Membership Inference Attacks (MIA) in AI models, introducing LUMIA—a framework that leverages linear probing to analyze unimodal and multimodal internal LLM states for improved attack detection and mitigation. Finally, I will close with future ideas and lines of research I would like to continue working on.
