Identifying indirectly related policies and data categories (beta)

AI Governance automatically shows the policies that apply to, and data categories that categorize, assets related to your AI use cases and the AI models they use. Showing these "indirectly related" policies and data categories on AI Use Case and AI Model asset pages allows you to quickly identify:

• Which policies (regulations and standards) apply to the Data Sets, Tables, and Technology assets associated with your AI use cases and AI models.
• If there is Personal Information (PI), Personally identifiable information (PII), or other sensitive information associated with your AI use cases and AI models.

AI Governance monitors and notifies you of changes to governed and categorized assets that are related to your AI Use Case and AI Model assets. For complete information, go to Email notifications for metadata changes.

Important We recommend that you not apply view permissions on the communities and domains that contain Policy and Data Category assets. Policy and Data Category assets on AI Use Case and AI Model asset pages are not visible to users who are restricted by view permissions.

In this topic

• Identifying indirectly related policies and data categories (beta)
• • Indirectly related policies
  • Indirectly related data categories
  • Enabling and disabling indirectly related policies and data categories
  • Limitations

Indirectly related policies

Policy Manager is used to define policies, regulations and standards that apply to your organization. Policies can be direct translations of official laws and regulations, rules that are defined by your organization, or customized versions of official laws and regulations that are adapted to fit your organization's profile, culture or processes.

In practice, you create Policy, Regulation, and Standard assets to capture all of the relevant details of a particular policy. All such assets are shown on the Policy Manager landing page. You can then create relations between Regulation assets and Standard assets, and assets of other types, to indicate that a specific asset is subject to a specific policy.

Policy-applied data that is associated with your AI use cases and AI models

As shown in the following image, when documenting an AI use case, the out-of-the-box Data and AI Models assessment prompts you to specify any inference data, training data, and output data associated with your AI use case.

Inference data is represented in Collibra as Data Set assets, Table assets, and Technology assets. The same is true of training data and output data.

The Data Set assets, Table assets, and Technology Assets that you specify might be governed by one or more policies – more precisely, via a policy's associated rules and standards. The ability to quickly identify all such indirectly related policies can be helpful. Fortunately, you don't have to go looking for related regulations and standards and linking them to your AI use cases and AI models. It's done automatically.

If the Data Set assets, Table assets, and Technology Assets that you link to your AI use case are governed by policies, the relevant Policy assets are automatically shown in the Other Policies That Apply widget, in the Legal, Ethics and Compliance section of the AI Use Case asset page.

Similarly, for AI Model assets, if you have policy-applied Data Set assets, Table assets, and Technology Assets that are related to an AI Model asset:

• The relevant Policy assets are automatically shown in the Other Policies That Apply widget, in the Related Content section of the AI Model asset page.
• If that AI model is related to an AI use case, the Policy assets are also shown on the AI Use Case asset page.

Indirectly related policies: applicable relation types

Not all policy-applied assets that are related to an AI use case or AI model will result in the Policy asset being shown on the AI Use Case or AI Model asset pages. Only the relation types shown in the following tables are applicable.

Assets related to AI Use Case assets

Relation type	Head role / corole tail	Public ID
uses AI Model	AI Use Case uses / used in AI Model	AIUseCaseUsesAIModel
infers from Asset	AI Use Case infers from / used to infer Asset	AIUseCaseTransformsAsset
trained by Asset	AI Use Case trained by / trains in Asset	AIUseCaseTrainedByAsset
has output Asset	AI Use Case has output / is output Asset	AIUseCaseHasOutputAsset

Assets related to AI Model assets (that are related to AI Use Case assets)

Relation type	Head role / corole tail	UUID
trained by Asset	AI Model trained by / trains Asset	AIModelTrainedByAsset
infers from Asset	AI Model infers from / used to infer Asset	AIModelInfersFromAsset
has output Asset	AI Model has output / is output Asset	AIModelHasOutputAsset

Indirectly related data categories

AI Governance automatically shows indirectly related Data Category assets on the AI Use Case and AI Model asset pages. This allows you to quickly identify if there are any Personal Information (PI), Personally identifiable information (PII), or other sensitive information associated with your AI use cases and AI models.

If the Data Set assets or Table assets that you link to your AI use case are categorized by data categories, the relevant Data Category assets are automatically shown in the Privacy Context widget, on the AI Use Case asset page.

The same is true for AI Model assets. If you have policy-applied Data Set assets or Table assets that are related to an AI Model asset:

• The relevant Data Category assets are automatically shown in the Privacy Context widget, on the AI Model asset page.
• If that AI model is related to an AI use case, the Data Category assets are also shown on the AI Use Case asset page.

Applicable relation types for indirectly related data categories

Not all Data Category assets that are indirectly related to an AI use case or AI model are shown on the AI Use Case or AI Model asset pages. Only the relation types shown in the following tables are applicable.

Assets related to AI Use Case assets

Relation type	Head role / corole tail	UUID
uses AI Model	AI Use Case uses / used in AI Model	AIUseCaseUsesAIModel
infers from Asset	AI Use Case infers from / used to infer Asset	AIUseCaseTransformsAsset
trained by Asset	AI Use Case trained by / trains in Asset	AIUseCaseTrainedByAsset
has output Asset	AI Use Case has output / is output Asset	AIUseCaseHasOutputAsset

Assets related to AI Model assets (that are related to AI Use Case assets)

Relation type	Head role / corole tail	UUID
trained by Asset	AI Model trained by / trains Asset	AIModelTrainedByAsset
infers from Asset	AI Model infers from / used to infer Asset	AIModelInfersFromAsset
has output Asset	AI Model has output / is output Asset	AIModelHasOutputAsset

Enabling and disabling indirectly related policies and data categories

You have to enable the AI Governance context option. If this option is not enabled:

• The Other Policies That Apply widget is not shown on AI Use Case and AI Model asset pages.
• The Privacy Context widget is shown on AI Use Case and AI Model asset pages, but it will be empty.

Note Keep in mind that this feature is in beta testing. Feel free to enable it in your testing and production environments to try it out.

Prerequisites

You have the ADMIN or SUPER role in Collibra Console.
You have a global role that has the System administration global permission.
The Services Configuration tab is available in the Collibra settings.

Steps

1. Open the Services Configuration page.

   1. On the main toolbar, click → Settings.
      The Collibra settings page opens.
   2. Click Services Configuration.
   3. Click Edit configuration.

   Open the DGC service settings for editing:

   1. Open Collibra Console.
      Collibra Console opens with the Infrastructure page.
   2. In the tab pane, expand an environment to show its services.
   3. In the tab pane, click the Data Governance Center service of that environment.
   4. Click Configuration.
   5. Click Edit configuration.
2. In the AI Governance Configuration section, enter the required information:
   

   Setting

   Description

   AI Governance context

   True: Show indirectly related policies and data categories on AI Use Case and AI Model asset pages. If changes are made to linked metadata, an email notification is sent to relevant users. Ensure that individual user preferences are appropriately configured in Collibra to receive email notifications. False (default): This feature is disabled. The Other Policies That Apply widget is not shown on AI Use Case and AI Model asset pages, and no email notifications are sent.

3. Click Save all.

Limitations

The following limitations apply to both indirectly related policies and data categories.

• Up to 1,000 indirectly related Policy assets, and up to 1,000 indirectly related Data Category assets, can be shown in the respective widgets on an AI Use Case or AI Model asset page. If there are more than 1,000 Policy assets, or more than 1,000 Data Category assets, only the first 1,000 assets are shown in the respective widgets.
• Changes to the relations between applicable assets are cached, and the widgets are updated approximately every hour. Therefore, if a relation between a data category and an applicable asset is added or deleted, for example, it can take approximately one hour before the Privacy Context widget reflects the change.

These limitations will be improved in a future version of Collibra.