Skip to main content

Prompt Guard Settings

On the Prompt Guard Settings page in your Pangea User Console , you can adjust the service configuration for optimal effectiveness and performance, as well as control the enablement of its Activity Log.

Analyzers configuration on the Prompt Guard General settings page in the Pangea User Console

Activity Log​

The Activity Log for Prompt Guard captures the details of service calls. You can view a summary of Prompt Guard activity on its Overview page and access individual logs on the Activity Log page in your Pangea User Console.

Learn more about using this feature on the Activity Log documentation page.

Configuration Options​

The Activity Log settings allow integration with the Secure Audit Log service, enabling attribution and accountability in your AI application.

Enabled by default, the integration uses the AI Activity Audit Log Schema, which is specifically designed to track AI application activity. When you click on the Activity Log tile, the dialog on the right will display the schema name and configuration ID. Prompt Guard logs its activity within this schema, capturing key details of each service call, including timestamps, input, output, detections, and contextual information.

Configuration options:

  • Enable Log Prompt Guard Activities (default).
  • Disable Log Prompt Guard Activities.

Audit Log Schema​

View and configure the schema fields​

You can view and configure the visibility of fields in your Prompt Guard audit schema, as well as set whether they are required on the Secure Audit Dialog configuration pages:

  1. In your Pangea User Console , click on the Secure Audit Log service name. If the service is not active, follow the wizard and accept the default settings to enable it.
  2. Select your AI Activity Audit Log Schema from the top left in the service sidebar.
  3. Click General in the sidebar.
  4. Under Secure Audit Log Settings, click the Audit Log Schema tile.
  5. In the dialog on the right, you can view the schema fields, update their visibility, and change whether they are required.
  6. Click Save to apply your changes.

Audit Schema settings on the Secure Audit Log General settings page in the Pangea User Console

Use the Schema in Your Application​

Optionally, you can use this schema to capture additional details in your AI application. This allows you to view your application-specific events alongside the service activity in the Prompt Guard Activity Log, with a summary on the service's Overview page.

To send log data to Secure Audit Log from your application code:

  1. Click the Overview link in the Secure Audit Log sidebar.
  2. On the Secure Audit Log Overview page, in the Configuration Details section, obtain the necessary credentials to make an API call to Secure Audit Log:
    1. Your AI Activity Audit Log Schema Config ID (ensure the schema is selected in the top left sidebar).
    2. The project Domain.
    3. The Default Token to authorize your application requests to the Secure Audit Log service APIs.

Learn more about configuring Secure Audit Log in its documentation.

Benign Prompts​

This setting helps mitigate false positives (FP). If innocent prompts are incorrectly flagged as malicious, you can add them to the Benign Prompts list. The setting also allows you to set a similarity score for the list. When Prompt Guard processes a new prompt, it will compare it against the stored Benign Prompts. If a match is found within the similarity threshold, the prompt is considered benign, and Prompt Guard will return that verdict for the new prompt.

Configuration options:

  • Enable - Requires the Benign Prompts list to be populated.
  • Disable (default).
  • Adjust Similarity Threshold.
  • Add or delete benign prompt examples:
    • Use the + Prompt button to add a new prompt.
    • Use the - button in the Prompts list to delete.
    • Use the Save button to apply your changes.

Malicious Prompts​

This setting helps mitigate false negatives (FN). If malicious prompts are incorrectly flagged as benign, you can add them to the Malicious Prompts list. The setting also allows you to set a similarity score for the list. When Prompt Guard processes a new prompt, it will compare it against the stored Malicious Prompts. If a match is found within the similarity threshold, the prompt is considered malicious, and Prompt Guard will return that verdict for the new prompt.

Configuration options:

  • Enable - Requires the Malicious Prompts list to be populated.
  • Disable (default).
  • Adjust Similarity Threshold.
  • Add or delete malicious prompt examples:
    • Use the + Prompt button to add a new prompt.
    • Use the - button in the Prompts list to delete.
    • Use the Save button to apply your changes.

Analyzers​

Prompt Guard includes a collection of analyzers used to determine whether a given prompt is malicious or benign. The list of analyzers may evolve as we continue to improve Prompt Guard's capabilities. The Analyzers configuration page allows you to enable or disable any analyzers available in Prompt Guard. By default, several analyzers are enabled, offering an optimized configuration for both security and performance.

Was this article helpful?

Contact us