Frontier AI models are increasingly capable of discovering and exploiting software weaknesses, prompting calls for more defensive preparedness and responsible release strategies. In this session, we translate that urgency into a practical, security-first approach to LLM penetration testing for real enterprise deployments.You will learn how to scope an LLM pentest beyond the model itself, covering the full application attack surface . Attendees will leave with a clear mental model for threat modeling LLM applications, a repeatable test plan that engineering teams can implement without slowing delivery. The focus is on defensible outcomes, measurable risk reduction, and aligning security findings with business impact.
