Model Armor

This security service enhances the safety of AI applications by proactively screening prompts and responses for risks like sensitive data exposure and prompt injection. It allows organisations to define guardrails that either inspect or block potentially harmful content, ensuring responsible AI practices across enterprise tools.

The feature is supported at no additional cost for NotebookLM Enterprise and is available across Gemini Enterprise editions, including Business, Standard, Plus, and Frontline. It provides admins with centralised control over security policies for their AI-powered workflows.

To enable it, admins must first create a Model Armor template in the Google Cloud console. For NotebookLM Enterprise, the template is then linked to the application using a PATCH request to the Discovery Engine API, specifying the resource names for user prompt and response templates.