Scope and Safety Tuning

GenerativeAgent includes default safety and scope guardrails to keep conversations aligned with business needs and to ensure GenerativeAgent engages only in appropriate topics. These tools allow you to:

• Define custom categories for what’s considered “in-scope”
• Configure input safety categories for allowed customer messages
• Maintain default safety protections while adding organization-specific allowances

​

Customizing Safety Settings

Safety controls are available in the Agent Identity section of the CXP. After making any changes to these settings, test them using the Previewer before deploying to production. Add Topics or content types that are considered safe to discuss with GenerativeAgent as part of Input Guardrails. You can also add specific examples of customer messages that are allowed, along with context explaining why these inputs are safe. To define valid topics for GenerativeAgent:

1. Navigate to Agent Identity > Safety
2. Click “Add Category”
3. Enter a category name and description
4. Provide specific examples of acceptable topics/requests
   **Customer:** Hi, I sold some stock last month and I'm trying to figure out how much I'll owe in taxes. Can you help?
   **Agent:** Happy to help you think through it. I'll need to know a few things first: how long you held the shares, and your total taxable income for the year. Did you hold the stock for more than a year before selling?
   **Customer:** Some of it I held for about two years, but I also sold shares I'd only owned for about four months.
   
5. Give explanations for why these topics are safe to discuss with GenerativeAgent
6. Save your changes

The default input guardrails remain active even when you add custom categories. Your configurations help customize permissible interactions while maintaining core safety features.

​

Next Steps

After configuring scope and safety settings, you may want to explore: