Creating Prompt Inspection Policies

Creating Prompt Inspection Policies

Barracuda SecureEdge Manager lets you create Prompt Inspection policies to control analysis and filtering of AI prompts and responses. Each policy defines sources, destinations, and sensitivity thresholds to determine if AI interactions are unscanned, Scan & Log, or Scan & Enforce. Policies evaluate in priority order (top to bottom); the first match sets the action.

Policy selection is based only on the configured source and destination. Classifiers do not determine which policy is selected.

Before You Begin

  • You must have a valid Barracuda SecureEdge Premium Access subscription.

  • You must have administrator access to the SecureEdge workspace.

  • TLS inspection must be enabled and configured.

  • Users must enroll with the SecureEdge Access Agent and route through a SecureEdge point of presence.

Available Templates

You can create a policy from a pre-built template or start from scratch using the Custom Policy option.

Template

Description

Template

Description

Monitor All AI Activity

Visibility first. Scans every AI prompt and logs findings without blocking anything.

Block Malicious Prompts

Blocks jailbreak, prompt-injection and obfuscated or junk input to any AI tool.

Protect Sensitive Data in AI

Blocks prompts containing PII such as card and social security numbers before they reach an AI tool.

Safe Use for Schools

Acceptable-use filtering for K-12 and EDU: blocks toxic, harmful and age-inappropriate prompts.

Professional Advice Guardrails

Stops staff from using AI tools for medical, legal or financial advice. For regulated and client-facing businesses.

Brand and Workplace Conduct

Blocks toxic, political, religious and harassing content in AI use at work.

Monitor Consumer AI Web Chat

Discovery for browser-based AI chat, the most common paste-anything risk surface. Logs only.

Allow Trusted Enterprise AI

Exempts sanctioned, contracted AI API integrations from inspection. Place above other policies.

Lockdown — High Security

Maximum coverage for high-risk tenants. Every classifier on, fails closed on timeout or oversize prompts.

Create a Prompt Inspection Policy

  1. Go to Barracuda SecureEdge and log in with your existing Barracuda Cloud Control account.

  2. From the drop-down menu in the top menu bar, select the workspace you want to create a Prompt Inspection policy for.

  3. Go to Security > Prompt Inspection on the left, then select Policies.

  4. The Prompt Inspection Policies window opens. Suggested policy templates are displayed at the top of the page.

  5. Do one of the following:

    • To create a policy from a pre-built template, select a template from the Suggested Policy Templates section. Click Add.

    • To create a policy from scratch, click + Add Prompt Inspection Policy in the top-right corner and keep Custom Policy selected.

      add-pmt-pol.png

       

  6. Click + Add Prompt Inspection Policy.

  7. A Custom policy templates opens. Note: To resize the template window, click the double‑arrow icon at the top-left corner of the window. To make the template full screen, click the bidirectional arrow icon.

  8. Specify values for the following:

    • Name – Enter a unique name for the policy.

    • Description – Enter a brief description.

    • Action – Select the action to take when the policy matches traffic:

      • Do Not Scan – Matched traffic is not sent to Prompt Inspection.

      • Scan & Log – Matched traffic is sent to Prompt Inspection for scanning but is allowed regardless of the scan results. The results are logged.

      • Scan & Enforce –Matched traffic is sent to Prompt Inspection for scanning. If it contains malicious content, the traffic is blocked.

    • In the Source section, specify the following:

      • All Sources – Applies the policy to all users and networks. No further configuration is required.

      • IP/Network – Enter one or more IP address or network in CIDR notation, and click + Add.

      • User/Group – Select users or groups from the drop-down list, or type to search. Select All Users to match every enrolled user.

    • In the Destination section, specify the following:

      • All Destinations – Applies the policy to all destinations.

      • AI Tool – Select the AI tools the policy applies to.

        fin-prompt.png
    • In the Classifiers section, specify values for the following:

      • Prompt injection – Click to enable. The prompt injection filter is an AI security tool that detects jailbreak attempts, where someone tries to trick the AI into ignoring its safety rules or revealing its internal instructions.

      • PII scanning – Click to enable. When enabled, scan prompt contents for sensitive PII such as credit card or social security numbers.

      • Gibberish – Click to enable. The gibberish filter detects random characters, keyboard mashing, and encoded content, helping identify client bugs or deliberate obfuscation attempts intended to hide malicious content.

      • Toxicity – Click to disable. The toxicity filter is an AI-powered safety tool that reads text and detects harmful language. Think of it as a spam filter, but instead of catching unwanted emails, it catches harmful content.

      • Banned Topics – Enter specific topics to block and click + Add.

        • Use the Suggestions (such as Hate speech, Harassment/bullying, Self-harm, Criminal activity, Illegal drugs, Cyber attacks, Profanity, Medical advice, Legal advice, Financial advice, Political content, and Religious content) as a starting point. Click a suggestion to add it automatically.

          classifiers.png

           

    • In the Settings section, do one of the following:

      • Global settings used – Leave Global settings used disabled to inherit the workspace default settings. By default, policies use the global settings.

      • To override the global settings, turn on the Global settings used toggle. When enabled, the toggle changes to Global settings overridden, and you can configure the following settings:

        • Strict mode – Click to enable. Content is allowed if prompt token limit is exceeded or inline processing times out. Scan results may still be available in logs. Not recommended for high security policies.

        • Prompt token limit – Enter the maximum number of tokens to inspect.

        • Inline timeout – Select the maximum time allowed for inline inspection from the drop-down menu (5–30 seconds).

          setting-PI.png

           

  9. Click Save. To clear every field and start again, click Clear all.

  10. After the configuration is complete, you can see that the new rule has been added to the Prompt Inspection policies table.

pmpt-ins-page.png

 

Edit a Prompt Inspection Policy 

  1. Select the workspace containing your Prompt inspection policies.

  2. Go to the Security > Prompt Inspection menu on the left and select Policies.

  3. The Prompt Inspection policies window opens.

  4. In the policies table, click the three-vertical-dots icon on the row of the policy you want to edit.

    edit-pol.png
  5. Click Edit policy. Edit the value you are interested in. 

  6. Click Save.

Note that the template drop-down does not appear when editing an existing policy.

 

Delete a Prompt Inspection Policy 

  1. Select the workspace containing your Prompt Inspection policies.

  2. Go to the Security > Prompt Inspection menu on the left and select Policies.

  3. The Prompt Inspection policies window opens. In the policies table, click the three-vertical-dots icon on the row of the policy you want to delete.

  4. Click Delete policy.

  5. The Delete Prompt Inspection policy <Name of policy>  window opens.

  6. Confirm the deletion and click Ok.

After deletion, the remaining policies retain their relative order and the deleted policy no longer applies.

 

Duplicate a Prompt Inspection Policy

  1. Select the workspace containing your Prompt Inspection policies.

  2. Go to the Security > Prompt Inspection menu on the left and select Policies.

  3. The Prompt Inspection policies window opens. In the policies table, click the three-vertical-dots icon on the row of the policy you want to duplicate.

  4. Click Duplicate policy.

  5. The create form opens with all fields prefilled to match the source policy, except the name.

  6. Enter a new name, adjust any fields you want to change, and click Save.

 

Reorder Prompt Inspection Policies

Policy priority is determined by the row order in the Prompt Inspection policies table. Use the drag handle to change priority.

  1. From the drop-down menu, select the workspace containing your Prompt Inspection policies.

  2. Go to Security > Prompt Inspection > Policies.

  3. In the policies table, click and hold the drag handle (= icon) on the left of the policy row you want to move.

  4. Drag the row to the desired position and release. The new policy order is saved automatically.

 

(Optional) Stop Prompt Inspection Scanning for a Workspace

To stop prompt inspection scanning while preserving your configuration, create a policy with the Do Not Scan action that matches All Sources and All Destinations, and place it at the top of the policy list. This preserves existing policies and their order while preventing traffic from being sent to prompt inspection.

Alternatively, you can delete all policies. When no policies exist, the Default action of Do Not Scan is applied to all traffic.

After the changes take effect, traffic is no longer scanned and no new prompt audit log entries are generated. 

 

Step 2. View in Prompt Audit Logs

You can view and verify prompt inspection policies enforcement events in Prompt Audit logs under Reports > Prompt Audit Logs in SecureEdge Manager. You can use filters to group and review log contents.

For more information, see Verifying Prompt Inspection Events in Prompt Audit Logs.

 

Further Information


We value your feedback.
If you have questions, suggestions, or feedback on our documentation, contact the Campus Product Documentation team.
For general product inquiries or technical support, please contact the global Barracuda Support team.