Sensitive data protection
Set which supported text patterns VeriProof masks before verified model calls and in session details. These settings apply across the organization. Start with the privacy requirement, then test the patterns your applications use.
Set the protection policy
Administrators, Governance Engineers, and Compliance Officers can configure:
- Whether redaction is enabled.
- The minimum sensitivity that triggers redaction.
- The policy applied to sensitive content. Permission to reveal personally identifiable information remains Administrator-only.
- Whether a reveal requires a written justification.
- Supported pattern checks, including email addresses, North American phone numbers, Social Security numbers, Canadian Social Insurance Numbers, valid payment card numbers, valid IBANs, IP addresses, and API keys or tokens.
- Canadian health and financial identifier checks for labelled text and recognized structured fields.
- Up to 50 exact application field names whose complete string values must be masked.
- Label, stars, or erase as the replacement style.
- Portable exact-match and regular-expression customer rules.
Review the selected patterns and replacement style before saving. Use invented values in tests, never actual customer identifiers or credentials.
Review the effective policy
Before changing a value, review Current policy summary and its last-updated time. Confirm:
- Whether redaction is enabled.
- The lowest sensitivity level that will be masked.
- That PII reveal remains restricted to Administrators.
- Whether a written reason is required before revealing content.
Change the policy safely
- Choose the minimum sensitivity level that matches the organization’s data-handling policy.
- Require a written reason when the organization needs an accountable record of reveal activity.
- Choose the supported sensitive-value patterns that connected SDKs should remove.
- For a Canadian application, select Use Canadian identifiers and add the exact health or financial field names used by the application.
- Select Save changes, then refresh the page and confirm the current policy summary.
A saved policy does not prove that model traffic is protected. Pattern changes take effect after a connected SDK refreshes its settings. They do not rewrite session details already stored in VeriProof. Verify the result with new synthetic data and a fake provider before relying on the change.
Understand what is covered
Sensitive Data Protection covers supported text that passes through the direct sanitizer or a verified adapter. It does not detect all protected health information (PHI). Names, street addresses, images, audio, and arbitrary documents are outside its inspection scope.
Postal codes are not broadly matched because ZIP+4 values can resemble Social Security numbers. Add an exact field name only when the whole value must be masked.
Understand reveal access
Redaction controls what is masked by default. Only Administrators have permission to reveal PII, and the redaction policy cannot grant that permission to another role. When a reason is required, the Administrator must enter it before access is granted. VeriProof records reveal activity for later review.