Risk Aggregation of Related AI Records

Risk Aggregation of Related Records enables systemic risk visibility across linked inventory objects in AI Governance. When enabled, parent objects (such as AI Systems or Projects) inherit risks from related child entities (such as Models or Datasets), allowing organizations to automate comprehensive risk views.

By default, risks are inherited automatically. In future iterations, users can choose whether to inherit risks at the point a relationship is created.

Things to Know

• Risk inheritance is on by default for supported relationships.

• There is no setting to control inheritance behavior.

• Manually unlinked risks may reappear if the child entity is updated.

• Specific risk suppression is not supported.

• Your tenant must be migrated to the latest Risk Scoring Methodology. For more information, see Risk Aggregation Upgrade Guide.

An AI system is built using multiple models and datasets. The models handle different functions like intent classification, answer generation, and content moderation. The datasets provide the training and reference material, such as historical support tickets, FAQs, and moderation examples. Each of these components (AI system, models, and datasets) carries its own risks. When relationships between the components are established, the risks from the models and datasets are linked directly to the AI system. This allows you to aggregate an overall AI system risk profile as a combination of system-level, model-level, and dataset-level risks.

The following seeded relationship types support risk aggregation:

• Managing AI Governance Relationships

• Risk Aggregation Scoring in AI Inventories

• Risk Scoring Upgrade Guide