Creating and Managing AI Models

Important:

The GenAI Component Intelligence for AI Models feature is under Public Preview. Please contact OneTrust Support to request early access. For the generally available model creation flow, see Managing Models in AI Governance.

The GenAI Component Intelligence for AI Models in OneTrust AI Governance provides a guided way to create a new model from the library, starting with seeded third-party AI model information rather than a blank form. It helps you add model details faster, review model information before selection, and link the model to existing inventory records.

About responsible use metrics

Responsible use metrics help assess fairness and potential disparities in outcomes produced by AI models. There is no single, universally accepted metric for summarizing fairness as one overall score. Today, OneTrust collects Responsible use metrics and will continue to expand them over time to produce a Model Fairness report.

Responsible use metrics complement the quality, safety, and relevance metrics, as well as other evaluation metrics, imported from supported AI platforms. For more information about capturing and reviewing those metrics, see AI Governance Agent and Model Evaluation Metrics.

Field Metric

Description

Demographic parity difference

This measures the difference in outcomes between groups.

A score of 0 indicates perfect demographic parity, while higher scores indicate more disparity between groups.
A score of 0.15 suggests moderate disparity.

Demographic parity ratio

This is the ratio of outcomes between groups.

A ratio of 1 indicates perfect parity, while lower ratios indicate more disparity.
A ratio of 0.8 suggests moderate disparity.

Equalized odds difference

This measures the difference in false-positive and false-negative rates between groups.

A score of 0 is perfect, while higher scores indicate more unequal odds.
A score of 0.1 suggests a small level of unequal odds.

Equalized odds ratio

This is the ratio of false-positive and false-negative rates between groups.

A ratio of 1 is perfect, while lower ratios indicate more unequal odds.
A ratio of 0.7 indicates moderately unequal odds.

To create a model from the model library

On the AI Governance menu, select Inventory > Models.

The Models screen appears.
Click the Add model button.

The Model library modal appears.
Click the Select button on the model card you want to use.

The Add new model modal appears.

You can also click the View details button to see more information about the selected model, including its capabilities, limitations, and safety and usage considerations.

Note:
To create a custom model instead, click the Create custom button on the Custom card, then complete the same steps in this procedure. Unlike a model selected from the model library, a custom model does not pre-fill the Provider and Model fields, so you must enter those values manually.
Enter the unique name for your model. The Provider and Model fields identify the selected library model and are pre-filled for you.
Select Add and review additional details (optional) to enter additional details and context for the model.
Field

Description

Owner

Choose from the available users at the organization level.

Description

Enter a short-hand description of the model's intended use.

Internal or 3rd party
Choose from the options:
- Self-managed open source software
- Third-party managed open source software
- Self-managed proprietary
- Third-party managed proprietary
Autonomy level

Choose the degree to which the model can perform tasks without human intervention.

Reference link

Enter the link to the source of model metadata or external documentation.

Model ID

Enter the model ID.

Record admin

Select the admin for the record.

Biases

Enter any potential biases that may impact the model.

Limitations

Specify any limitations that might impact the model's operation.

Demographic parity difference

Enter the difference in positive outcomes between groups.

Demographic parity ratio

Enter the ratio of positive outcomes between groups.

Equalized odds ratio

Enter the ratio of prediction error rates between groups.

Generative AI

Choose whether the model is generative or not.

Framework

Choose from the available framework options.

Hosting provider

Choose the cloud provider hosting the model from the available options.

Model version

Enter the model version.

Programming language

Choose from the available programming language options.

Type of learning

Choose from the following: Supervised, Unsupervised, or Reinforcement.

Task type

Choose which task this model is intended to complete in your governance program.

Is master

Specify whether the record is a master record.

Master

Select the master record linked to this record.
Click the Continue button.

The Add relationships screen appears.
You can add relationships to link the new model to an existing inventory in your tenant. Define the relationship type, target type, and target record, then click the Add button.
After you add relationships, click the Continue button.

The Review screen appears.
Review the details for the new model.
Click the Create model button. The model is created and added to your model inventory.

To review the model deprecation status

If a model in the Model library is deprecated, a warning appears for that model and for its associated model records.

On the AI Governance menu, select Inventory > Models.

The Models screen appears.
In the Model overview tab, click a Warning icon on the impacted model card.

The warning message appears.
Click the View all impacted records link.

The model side pane appears. In the Models tab, you can view all model records for the selected model.
Click the model record you want to review.

The deprecation status for the associated model appears in the model details.

To change or remove the associated model from the model record, see To manage model details.

To manage model details

On the AI Governance menu, select Inventory > Models.

The Models screen appears.
Go to the Model records tab and select a model.

The Inventory details screen appears with the Details tab open by default.
In the Details tab, update the model details as needed, and then click the Save button.
In the Model information tile, you can review, change, or remove the associated model.
- To review more information, click the View details button. The model details panel appears.
- To update the association, click the Context Menu icon, then select Change model.
  
  Click the Select button for the desired model.
- To remove the association, click the Context Menu icon, then select Remove model.
  
  Click the Remove model button on the confirmation modal.

To manage a model workflow

On the AI Governance menu, select Inventory > Models.

The Models screen appears.
Go to the Model records tab and select a model.

The Inventory details screen appears.
Click the Next stage button to move the model to the next workflow stage.
To revert workflow stage, click the stage that you want to return to.

The Revert stage modal appears.
Click the Continue button to revert the workflow. Optional: run rules and actions associated with this stage transition.
To advance the model to a later stage, click the stage that you want to move to.

The Skip workflow stage modal appears.
Click the Continue button to advance the workflow. If any of the skipped stages are approval stages, approvers will not be added or contacted.

To delete a model

On the AI Governance menu, select Inventory > Models.

The Models screen appears.
Go to the Model records tab and click the Context Menu icon on the model you want to delete.
Select Delete.

The Delete record modal appears.
Click the Confirm button. The model is deleted.

AI Governance Manager