Keynote | Agentic AI: Understanding and Securing the Next Frontier of Intelligent Systems

Imagine giving your intern not just access to your computer, but also your credit card, API keys, and permission to make decisions on your behalf, that's essentially what we're doing with agentic AI. Unlike traditional AI that focus on classification predicting outcomes, these systems actively take actions, use tools, and pursue goals with minimal human oversight. This creates complex new attack vectors: agents can be tricked into transferring funds, manipulated through crafted inputs, sent astray by misinformation and hallucinations from another agent in the chain, or develop unexpected behaviors that sidestep our security assumptions. We'll dive into real-world threat scenarios, explore why traditional security models fall short, and discuss practical approaches for containing and governing the next frontier of intelligent systems.