What happens when your AI agent encounters unexpected, adversarial, or incomplete inputs? In this session, we explore how to proactively “attack” your own AI systems to make them more robust. We will cover practical red teaming techniques including adversarial prompting, context manipulation, and tool exploitation. Using a hands-on demo in Microsoft Foundry, we will simulate edge cases and stress test an AI agent. This session is designed for practitioners looking to move from experimentation to production-grade AI systems with built-in resilience. Key Takeaways - Simulate real-world adversarial scenarios and edge cases - Stress test agents across context, tools, and decision flows - Identify risks in tool usage, and memory handling - Apply practical guardrails and monitoring using Microsoft Foundry
