Sanitize Prompts and Responses with Model Armor

Why it matters

Model Armor is an enforcement boundary for prompts and model responses. The certification expects you to place it deliberately among identity, approval, tool, and data controls rather than treat it as a complete security system.

What to practise

Enable the service, create a template, inspect user prompts and model responses, and exercise checks for unsafe content, malicious URIs, prompt injection, and sensitive data.

How it fits

Use the exercise after subsection 5.2’s threat-model and guardrail Bites. Observe which risks the filter can detect and which still require authorization, approval, isolation, or application logic.