As large language models (LLMs) gain the ability to act, browse, automate, and interact with real-world systems via the Model Context Protocol (MCP), they also expose new and unpredictable attack surfaces.

In this talk, I’ll introduce MCP — the protocol powering tool-using agents — and walk through the emerging security threats it brings, including tool injection, prompt exploits, session hijacking, and remote code execution. We’ll explore practical, field-tested defenses and governance strategies that can help teams build AI-enabled systems that are powerful and safe.

Whether you're a developer integrating LLMs or a manager responsible for shipping secure AI products, this session will equip you with the mental models, examples, and frameworks to secure your agentic architectures.

Key Takeaways
Understand MCP: What the Model Context Protocol is, how it works, and why it matters in agentic AI systems.
Recognize Security Risks: Learn the top threats including tool injection, session hijacking, and RCE — with real-world inspired examples.
Apply Defensive Design: Discover actionable mitigations like tool whitelisting, RBAC, sandboxing, and red teaming for AI workflows.
Go Beyond the Model: See why prompt injection and data leakage aren’t just "prompt problems" but architectural concerns.
Stay Ahead: Get a curated resource list of blogs, OWASP guidance, and security tools for AI risk management.