Model Armor is a Google Cloud service designed to enhance the securityand safety of your AI applications, particularly those using Large LanguageModels (LLMs). It works by inspecting the prompts sent to your models and theresponses generated by them, helping you mitigate risks and enforce responsibleAI practices.

Configure templates

Define how Model Armor should screen content by creating andusing Model Armor templates. A template is a reusable configurationset where you specify which filters to enable, the confidence levels for thefilters, and the enforcement type for each filter. For more information, seeCreate and manage templates.

Configure floor settings

To ensure a baseline level of protection, security administrators can configurefloor settings at the organization, folder, or project level. These settingsmandate minimum filter requirements that all Model Armor templatescreated within that scope must adhere to, helping to prevent overly permissiveconfigurations. For more information, seeConfigure floor settings.

Sanitize prompts and responses

When a user submits a prompt to your application, your application first sendsthis prompt to Model Armor. Model Armor processesthe prompt through the enabled filters in the template and returns a responseindicating whether any policy violations were found and detailed results fromeach filter. Your application logic then decides what to do next.

When an LLM generates a response, before displaying this response to the user, yourapplication sends it to Model Armor. Model Armorscreens the LLM output using the filter configurations defined in the templateand returns the analysis results. Your application then decides whether to showthe response to the user, potentially blocking it if violations are found.

For more information, seeSanitize prompts and responses.

What's next

• Model Armor overview