To help provide you a secure and reliable experience, Gemini Enterprise – Business edition includes built-in AI safety and security safeguards. These protections can be used to proactively screen your queries and uploaded files, and the assistant's responses to help you mitigate the potential for harmful content, malicious attempts to manipulate the AI, and sensitive data leakages.
These safeguards are a core part of the service. Features like Model Armor provide this protection, which is included in your subscription at no additional cost.
Automated safety protections
The system provides automated safeguards designed to address several types of potential security and safety risks:
- Harmful content: Safeguards can identify and block responses or queries that may contain unsafe or harmful material.
- System manipulation: The screening process helps reduce the risk of "prompt injection" or other attempts to bypass safety filters.
- Data leakage prevention mechanisms: Built-in filters scan for sensitive information to help prevent accidental disclosures.
How your data is screened
Every interaction undergoes a multi-step screening process:
- Input sanitization: Before your request is processed by the underlying model, the system scans your prompt and any attached files.
- Response sanitization: Before a response is displayed to you, the generated text is scanned for potential safety or policy violations.
- Automatic blocking: If a violation is detected based on default safety templates, the system automatically blocks the request or response.
Built-in management and maintenance
These security features are designed to work seamlessly without requiring manual configuration:
- Managed by Google: This protection is powered by pre-configured safety templates managed and updated by Google.
- No setup required: Unlike other Gemini Enterprise editions that may require administrator configuration, these protections are active for all users by default.
- Continuous updates: Google continuously maintains these safeguards to address new and evolving AI safety risks.