AI agents are becoming active participants in enterprise systems—calling APIs, accessing data, invoking MCP tools, and taking actions on behalf of users. That creates a new security boundary. This session presents a zero-trust architecture for securing agentic applications on Azure. We will examine agent identity, least-privilege authorization, managed identities, secrets, tool permissions, MCP security, prompt injection, approval gates, and auditability. Through practical architecture patterns and attack scenarios, attendees will learn how to prevent an AI agent from becoming an overly privileged automation layer while still enabling useful autonomous workflows.

Focus: Securing autonomous AI agents, MCP tools, identities, permissions, APIs, secrets, and privileged actions using zero-trust architecture.