LLM Security Through the Offensive Lens

As organizations rush to adopt Large Language Models (LLMs), adversaries are equally quick to probe their weaknesses. This session draws on real-world offensive security testing of LLM applications, exposing risks such as unbounded consumption, excessive agent permissions, and prompt injection. Through the case studies, we will share how these flaws were uncovered and exploited and share key lessons security teams can apply to strengthen their defences and secure LLM adoptions.