Slide 1 · Title
Setting the Scene
The room is full of security professionals. The speaker opens with a show-of-hands question: "How many of you last year were asked to secure the AI features being added to your apps?" Most hands go up. Then: "And how many of you, before you'd even finished doing that, were asked to now secure the AI agents?" Almost every hand stays up - and a ripple of knowing laughter moves through the audience.
That exchange captures exactly where the industry sits in 2026. The pace of change in AI has been so rapid that security teams are routinely handed a new problem before they have had a chance to solve the previous one. Features became copilots. Copilots became agents. Agents became multi-agent systems running autonomously in production pipelines. And through all of it, the security profession has had to keep pace.
The Promise of This Talk
This session is not about theory. It is not about distant, hypothetical risks. It is about what is happening right now, in real organisations, and what you can do about it this week.
The promise is simple: leave with a clear, actionable framework for building and deploying AI agents responsibly - one that maps to the security fundamentals you already know, extended for the agentic world.
Why Security Professionals Are the Most Important People in the Room
There is a temptation in fast-moving technology cycles to feel like security is always catching up. But here is the truth: you are not behind. The fundamentals of threat modelling, least privilege, identity management, observability, and incident response are not obsolete - they are more valuable than ever, because every single one of them applies directly to securing AI agents.
What changes is the attack surface, the autonomy, and the speed. What does not change is the discipline required to get it right. This talk is about applying that discipline to a new problem. You already have everything you need.