Learn to defend LLM and agent workflows using different types of guardrails via Meta's llamafirewall

Explore Meta's Llama Firewall, a powerful collection of guardrails designed to protect LLM and agentic workflows. Learn how three different types of guardrails and safety measures: Prompt Guard, Code Shield, and Alignment Check work together to mitigate security risks like jailbreaks and insecure code. We also walk through a live demo and discuss how to build effective mental security models for AI. Let me know additional questions in the comments :) Related links: 13:32 - Analyzing the Agent Workflow 15:30 - Running the Live Demo 17:14 - Testing Security Attacks 18:22 - Code Shield and Alignment Examples 19:32 - Building a Security Model Guardrails meta post: https://blog.kjamistan.com/ai-guardra... LlamaFirewall home page: https://meta-llama.github.io/PurpleLl... OWASP intro to Static Code Analysis: https://owasp.org/www-community/contr... Semgrep Community Edition: https://semgrep.dev/products/communit... How to install and setup environment variables and the built-in test script I used: Where to edit for using a local or other LLM: https://github.com/meta-llama/PurpleL... 00:00 - Introduction to Llama Firewall 01:14 - Breakdown of Guardrail Types 02:04 - Prompt Guard and Alignment 04:16 - Importance of Task-Specific Models 06:06 - Exploring Code Shield 08:52 - How Alignment Check Works 11:32 - Implementing the Firewall