Masterclass "Safe AI: Discovering and Resolving Vulnerabilities in Models and Chatbots"

A practical intervention on how to translate risk categories into concrete tests, using open-source and specialized tools. Together with Mauro Medda, Co-founder & CTO of HikmaAI, and Mauro Dore, AI Security Engineer at HikmaAI, we will see live demos with Garak (NVIDIA) and DeepTeam on three attack scenarios from the OWASP Top 10 for LLM Applications: prompt injection (LLM01), sensitive information disclosure (LLM06), and excessive agency (LLM08). We will then discuss the security of AI agents, focusing on skill analysis and evaluating risks related to MCP servers, tool access, and autonomous workflows, using HikmaAI's CLI for an integrated approach to testing agents, models, and infrastructure. Each scenario will be performed live, with measurable and immediately replicable results: an event designed for security engineers, developers, and red teams who want to move from theory to operational AI security. The masterclass takes place within Connessioni Future, an event by the Fondazione di Sardegna implemented by Innois. 📍Pavilion Tavolara, Sassari - Room 2🕘11:30Also sign up for the other days of Connessioni Future - October 6: https://luma.com/nmn8ew8d- October 7: https://luma.com/cljllhjs- October 8: https://luma.com/uyhg3aaxDiscover more about www.connessionifuture.it