Data Redaction - Cognigy Documentation

Data Redaction Overview

The Data Redaction feature provides an easy way to detect and redact any data, including personally identifiable information (PII), to ensure compliant logging and analytics. This feature is important in the context of conversational AI due to privacy, trust, compliance, and responsible AI practices.

Key Benefits

• Enhanced Data Handling. Ensures data protection during the processing, logging, and storage of conversations containing sensitive data. Compared to data masking, which hides or selectively excludes information at the Endpoint or Flow level ( Blind Mode Node), data redaction fully removes the sensitive data from all logs and analytics for stronger data protection.
• Easy Configuration. Includes a set of default data types you can select to redact, such as credit card numbers or email addresses.
• Customization. Lets you add custom data types to redact. You can use regular expressions (regex) to define patterns for custom data types to detect.

Prerequisites

For on-premises installations, you control whether data redaction is activated in the values.yaml file. By default, data redaction is activated. Depending on which Cognigy.AI version you use, you can deactivate data redaction as follows:

• Cognigy.AI 2026.9 and later
• Cognigy.AI 2026.8 and earlier

In the values.yaml file, configure the commonConfigs.cognigyRollout.pii-data-redaction.yaml key:

Deactivate data redaction for all organizations

To deactivate data redaction for all organizations, set percentage to 0.

commonConfigs:
  cognigyRollout:
    pii-data-redaction.yaml: |
      organisations:
        percentage: 0
        disabled: []

Deactivate data redaction for specific organizations

To deactivate data redaction for specific organizations, assign organization IDs to the disabled array.

commonConfigs:
  cognigyRollout:
    pii-data-redaction.yaml: |
      organisations:
        percentage: 100
        disabled:
          - "<organization-id-1>"
          - "<organization-id-2>"

In the values.yaml file, set the following environment variables:

• FEATURE_PII_DATA_REDACTION_ROLLOUT — the percentage of organizations for which data redaction is activated. Set it to "100" to activate data redaction for all organizations, or to "0" to disable it for all. Values in between activate the feature for that percentage of organizations. The system automatically and consistently determines which organizations are included.
• FEATURE_DISABLE_PII_DATA_REDACTION_WHITELIST — a comma-separated list of organization IDs to explicitly exclude from data redaction. This setting takes precedence over FEATURE_PII_DATA_REDACTION_ROLLOUT. For example, FEATURE_DISABLE_PII_DATA_REDACTION_WHITELIST="<organization-id-1>,<organization-id-2>".

Limitations

• Log entries with data payloads over 2.5 MB aren’t redacted. They’re replaced with a placeholder message. For on-premises installations, you can configure the maximum payload size using the PII_REDACTION_DEFAULT_MAX_DATA_SIZE_IN_MEGABYTES environment variable in values.yaml. Setting this value higher than 2.5 MB may cause performance issues. Increase it at your own risk.

Restrictions

• Data redaction uses regex patterns to detect and redact data, which can:
  • Cause false positive matches. For example, the credit card redaction pattern may cause false positives for 3- or 4-digit numbers that are sent in the message input but aren’t CVC or CVV numbers.
  • Skip patterns that don’t match the default types or a custom data type you define. For example, if the user enters a phone number with / or misspells a company name set to be redacted in a custom data type.
• Custom data types support only Google’s RE2 syntax library for regex patterns.
• Depending on the Cognigy.AI version you use, data redaction works differently with Slots:

Cognigy.AI 2026.3 and later versions

Data redaction works for patterns detected by Slots only if the Slot matches the entire data redaction pattern. If a Slot matches only part of the pattern, the data isn’t redacted as expected. For example, if a credit card number is split across multiple Slots, such as a number Slot and a date Slot, the credit card number isn’t fully redacted.

Cognigy.AI 2026.2 and earlier versions

Data redaction doesn’t work for patterns detected by Slots. To redact such data, deactivate Slots at the Flow or the Project level.

How to Use

To use the data redaction feature, first check the default data types. If none are suitable or if an existing type doesn’t match your data due to its pattern rules, you can create a custom data type.

Default Data Types

The data redaction feature includes the following default data types, covering common forms of PII:

To activate the data redaction for default data types, follow these steps:

1. In Manage > Settings, go to the Data Redaction Settings section.
2. Click the subsection of the data type you want to redact and configure the following:
   • Activate Redaction — toggle on to activate the redaction for the data type.
   • Redaction Pattern— select one of the following options:
     • Default — use the default redaction pattern. Each data type has a default redaction pattern, for example, the redaction pattern for credit card numbers is [CREDIT CARD REDACTED].
     • Custom — enter a custom redaction pattern. The custom redaction pattern should be enclosed in square brackets and must not contain spaces or special characters. For example, [PROTECTED_INFO].
3. Under Scope, select where the data type is redacted:
   • Logs — redact the data type on the Logs page.
   • Analytics — redact the data type in Insights and the Cognigy.AI OData endpoint.

Custom Data Type

If you need to redact data that isn’t included in the default data types, you can add custom data types. To add a custom data type, follow these steps:

1. In Manage > Settings, click the Data Redaction Settings section to expand it.
2. At the bottom of the section, click + Custom Data Type and configure the following:
   • Name — enter a unique name.
   • Regex — enter a regular expression to detect the data type pattern.
   • Custom redaction pattern — enter a custom redaction pattern. The custom redaction pattern should be enclosed in square brackets and must not contain spaces or special characters. For example, [PROTECTED_INFO].
3. Under Scope, select where the data type is redacted:
   • Logs — redact the data type on the Logs page.
   • Analytics — redact the data type in Insights and the Cognigy.AI OData endpoint.

Examples

The following table includes examples of regex patterns and redaction patterns for custom data types.

More Information

• Logs
• Insights
• Slots
• Project Settings