Skip to main content

Creating AI guardrails

You can create AI guardrails to ensure that the generative AI systems you use in Brightspot operate within your organization's boundaries.

To create an AI guardrail:

  1. From the left navigation, under Admin, click Sites & Settings.

  2. Under the Integrations tab, expand the AI cluster.

  3. Under Guardrail, select Create New.

  4. Select either Bedrock Manual Guardrail (manually configured within Brightspot) or Bedrock Imported Guardrail (already existing and ready to be imported) and do the following depending on your selection.

    • Bedrock Manual Guardrail—

      1. Under Guardrail Provider, select the guardrail provider that will validate usages of AI against the rules you set.

      2. Under Credentials, select one of the following:

        • Default—The default credentials configuration.
        • Assume Role—Assumes the AWS role of your choosing to enforce guardrails. Requires that you enter a value in the Role ARN field (the role you want Brightspot to assume when accessing the AI service) and the Session Name field (the label you want to assign to the session created when assuming the role for tracking purposes).
        • Static—Allows you to provided fixed AWS credentials instead of assuming the role dynamically. Requires that you enter a value in the Access Key ID field (identifies which AWS account is making the request), Secret Access Key field (used with the access key ID to make requests securely), and Secret field (a required additional secret).
      3. Under Region, retain the default (us-east-1) or enter a different region where your AI resources are located.

      4. Toggle on Enable for Create with AI to enable this guardrail for Brightspot's Create with AI feature.

      5. Toggle on Enable for Ask AI to enable this guardrail for Brightspot's Ask AI feature.

      6. Under Name, give the guardrail a name.

      7. Under Description, give the guardrail a description.

      8. Under Harmful Content Filters, select one of the following:

        • Hate Content Filter—Filters out input and output that can be characterized as hate speech.
        • Insults Content Filter—Filters out input and output that can be characterized as an insult.
        • Misconduct Content Filter—Filters out input and output that can be characterized as misconduct (criminal activity, harm).
        • Sexual Content Filter—Filters out input and output that can be characterized as being sexual in nature.
        • Violence Content Filter—Filters out input and output that can be characterized as violent.
      9. Under Denied Topics, click Add Denied Topic, then do the following:

        1. Under Name, give the denied topic a name.
        2. Under Description, add a description for the denied topic.
        3. Under Sample Phrases, add phrases that trigger the detection of the denied topic. Click to add additional phrases.
      10. Under Sensitive Information Types, select from a large list of sensitive personal identity information types.

      11. Click Save. After saving, Brightspot assigns a Guardrail Identifier and lists the Guardrail Version.

    • Bedrock Imported Guardrail—

      1. Under Guardrail Provider, select the guardrail provider that will validate usages of AI against the rules you set.

      2. Under Credentials, select one of the following:

        • Default—The default credentials configuration.

        • Assume Role—Assumes the AWS role of your choosing to enforce guardrails. Requires that you enter a value in the Role ARN field (the role you want Brightspot to assume when accessing the AI service) and the Session Name field (the label you want to assign to the session created when assuming the role for tracking purposes).

        • Static—Allows you to provided fixed AWS credentials instead of assuming the role dynamically. Requires that you enter a value in the Access Key ID field (identifies which AWS account is making the request), Secret Access Key field (used with the access key ID to make requests securely), and Secret field (a required additional secret).

      3. Under Region, retain the default (us-east-1) or enter a different region where your AI resources are located.

      4. Toggle on Enable for Create with AI to enable this guardrail for Brightspot's Create with AI feature.

      5. Toggle on Enable for Ask AI to enable this guardrail for Brightspot's Ask AI feature.

      6. Under Internal Name, give the guardrail a name.

      7. Under Guardrail Identifier, select an identifier from the list.

      8. Under Guardrail Version, select a version from the list.

      9. Click Save.