AI Red Teaming — Why & How to Jailbreak LLM Agents

🎥 Recorded live at the MLOps World | GenAI Summit 2025 — Austin, TX (October 9, 2025) Session Title: AI Red Teaming: Why & How to Jailbreak LLM Agents Speaker: Alex Combessie, Co-Founder & Co-CEO, Giskard Abstract: As AI agents become more capable, they also become more vulnerable. In this session, Alex Combessie from Giskard explores how adaptive, multi-turn attacks can “jailbreak” LLM agents — and what it takes to defend against them. This talk introduces the foundations of AI Red Teaming, highlighting how attackers exploit conversational consistency and short-term context to manipulate models. Attendees will see why static, one-off testing is no longer enough, and how continuous, automated red teaming with human-in-the-loop (HITL) monitoring can uncover and neutralize emerging security risks before they reach production. What you’ll learn: • How adaptive, multi-turn attacks compromise AI agents • Why static red teaming falls short against evolving threats • How attackers exploit LLMs’ reliance on context and conversational flow • Practical strategies for continuous red teaming and automated threat detection • How to integrate HITL oversight to ensure robust AI security.