Cisco's AI Defense Workshop covers how to use the Cisco 3-step framework—Discovery, Detection, and Protection—to secure the AI lifecycle. Participants will practice running algorithmic red teaming simulations, scanning model files for malicious payloads, and configuring runtime guardrails. The workshop also covers securing agentic workflows using the Model Context Protocol (MCP) to prevent goal hijacking and context poisoning.

What to expect

You will work through the workshop as a Senior Security Engineer at PseudoCo, a fictional organization deploying generative and agentic AI across its enterprise. The workshop covers four use cases and fifteen success criteria.

The workshop includes:

• AI asset discovery using Cisco AI Defense
• Algorithmic red teaming to test models for prompt injection, jailbreaking, and harmful content generation
• Model file scanning for embedded threats in .pkl, .h5, and .onnx files
• Runtime policy configuration to block PII, PHI, and PCI data leaks
• MCP server scanning for vulnerabilities and intent-based threats
• Agentic protection against tool poisoning, prompt injection, and rug pull attacks
• Side-by-side testing of Monitor and Enforce policy modes using the AI Defense Demo Runner

Who should attend

• Security engineers and architects
• AI and ML platform owners
• Application security teams
• IT and security leaders responsible for AI governance

What you will learn

• How Cisco AI Defense applies the Discovery, Detection, and Protection framework to AI systems
• How to validate models before deployment and protect them at runtime
• How to detect and block agentic AI threats specific to MCP-based architectures
• How to build runtime policies that enforce data protection and content safety