AI agents built on Azure AI Foundry introduce a new class of security challenges that traditional perimeter-based thinking does not solve. An agent that can call APIs, query databases, and invoke tools needs a trust model where every action is verified, every credential is scoped, and no component is implicitly trusted, even within your own Azure subscription.

This session walks through a zero trust architecture for production AI agent workloads on Azure AI Foundry. We cover Managed Identity for agent-to-service authentication without secrets, Azure RBAC scoping for tool and data access, network isolation with Private Endpoints and Azure Container Apps ingress policies, prompt injection defense patterns, and audit logging with Azure Monitor.

Real patterns from a production AI-powered application are used throughout, so attendees leave with implementation-ready guidance rather than theory.