Enable AI Guardian for AI agents

Identify and block offensive messages that are sent by human agents automatically by enabling AI Guardian in AI agents. With this capability, you can help reduce your agentic workflow or test from being exposed to harmful content.

Before you begin

Role required: admin

About this task

The AI Guardian, which is a ServiceNow AI Platform capability in the ServiceNow Otto panel, is a set of guardrails that are designed to intercept and mitigate offensive, sensitive, or security-related issues that may arise during interactions with the ServiceNow Otto application.

For example, let's say that AI Guardian detects an offensive message in the execution plan of an agentic workflow. When you try to trigger the plan or test it, AI Guardian can step in to terminate the plan or test because it detected harmful content at the first step of the execution plan.

For more information about the different guardrails, see AI Guardian.

Procedure

1. Configure Offensiveness for AI agents.
   1. In AI Agent Studio, navigate to the Settings tab.
      You’re directed to the Offensiveness page.

      

   2. Turn on the Offensiveness setting for AI agents by using the toggle button.

      

   3. Configure the detection impact by selecting the options icon () to enable the detection impact to use the following options:

      

      • Edit: Choose the detection impact between logging or both blocking and logging.
        Enable the detection impact:

        1. Select the Edit option.
        2. On the Detection impact page, select either the Log or Block and log button based on your requirements.
           Note:

           You can switch the detection impact from Log to Block and log or from Block and Log to Log at any time.
        3. Select Save.
      • Export: Exports the offensiveness-based logs as a .CSV file. Logs can be analyzed for the types of content that are being identified so you can take other preventive measures, such as changing conversational questions.
2. Configure Prompt Injection for AI agents.
   1. In the AI Agent Studio, navigate to Settings > Prompt Injection and select Configure.
      You’re directed to the AI Admin Hub page to configure the Prompt Injection.

      

      Note:

      For more information about configuring the Prompt Injection, see Configure prompt injection attack protection.

      When you configure the Prompt Injection for an agentic workflow by using the required instructions, the system is designed to detect harmful content and block the conversation.