Build Secure AI Agents - Microsoft's Stack, Your Innovation

Explore Microsoft's comprehensive approach to securing agentic AI systems throughout their entire lifecycle in this 46-minute conference talk from Microsoft Ignite 2025. Discover how to define agentic AI and understand autonomy design choices while learning about emerging threats including AI-augmented cyberattacks and user manipulation risks. Examine real-world scenarios through an employee expense agent threat modeling exercise and understand critical security challenges such as memory poisoning and data leakage risks. Learn strategies for managing agent attack surfaces and latency considerations, while exploring how to define proper agent schemas and establish explicit role boundaries. Master essential security practices including data sanitization, guardrails filtering for secure inputs, and implementing robust logging and monitoring systems for observability and attack preparedness. Gain insights from Microsoft Security and Microsoft Foundry experts as they demonstrate practical tools and design strategies aligned with Secure Future Initiative principles, concluding with governance and security implementation using Microsoft products. Presented by Neta Haiby, Harmony Mabrey, and Alyssa Ofstein, this advanced-level breakout session provides actionable guidance for building safe, resilient AI agents in enterprise environments.