Securing AI Agents: What could go wrong? A hands-on session for understainng threats and how to mitigate them

What could go wrong when an AI agent acts on your behalf, such as reading and using a Web?

In this interactive session, Threat Modeling Connect (TMC) will introduce its community and show how threat modeling helps us examine emerging AI technologies. Using agentic AI and WebMCP as a practical case, we'll share a few threats already identified in ongoing work, then invite participants to hands-on facilitated activities o challenge the model, find gaps, and discuss possible responses.

We'll close by looking at what evidence is needed to support AI risk management and, where applicable, EU AI Act compliance. No prior threat modeling experience is needed. Bring your perspective and see how you can contribute.