Hands-on playbook · 55 min
Guardrail Libraries Playbook: NeMo Guardrails Tested, Llama Guard 4, Prompt Guard 2 and LlamaFirewall in an Agent
A tested, hands-on guide to open-source LLM guardrails: a NeMo Guardrails 0.24.1 configuration exercised offline, including what its jailbreak rail does when the detector returns 503 (it lets the request through), plus Prompt Guard 2's limits, LlamaFirewall's scanners for agents, and a fail-closed policy to defend.
- A NeMo Guardrails configuration loaded and checked with check_async, no LLM needed
- Measured: a detector answering 503 lets prompts through; a refused connection blocks
- Self-check rails that refuse to load without a prompt, and typed exceptions
- Prompt Guard 2: benign or malicious, a 512-token window, and what it does not judge
- LlamaFirewall scanners for agents, including hidden Unicode tag characters
- A tested fail-closed decision function for combining several rails
- Server hardening, a 90-day rollout and an interview drill
- 01Under the hood
- 02Hands-on lab
- 03Recipes
- 04Rollout runbook
- 05Troubleshooting
- 06Interview drill
- 07Sources
Every worked answer in all 17 topics, every lesson in every course, the full concept curriculum, hands-on tool playbooks, the complete practice-test bank, and the downloadable security field guides.
6 months · One payment · No auto-renewal
An independent playbook for interview preparation and learning. AISecurityInterviews is not affiliated with or endorsed by NVIDIA (NeMo Guardrails); Meta (Llama Guard, Prompt Guard, LlamaFirewall); Protect AI (LLM Guard, archived), and product names are trademarks of their owners. Use these procedures only on systems you own or are authorised to test, and check the vendor's documentation before changing production.