What's new and changed in watsonx.governance

New query parameter for flagging historical records in the Datasets API and Python SDK

You can now use the is_historical_data query parameter to identify records as historical when uploading data through the Datasets API or Python SDK.

When you upload records using the POST /v2/data_sets/{data_set_id}/records REST endpoint or the store_records method in the Python SDK, set is_historical_data=true if your records include values in the scoring_timestamp or record_timestamp fields and are intended for historical evaluation. If you omit the parameter or set it to false, the records are used in current evaluations, even if the timestamps are in the past.

This enhancement improves clarity and control when managing timestamped data for model evaluation workflows.

Enhancements to Governance Console

Create child object from generative AI output

Users can now generate child objects, such as action items or controls, based on generative AI output. This feature streamlines content creation and helps users act on AI-generated insights more efficiently.

For more information, see Configuring model outputs in the IBM OpenPages documentation.

Customize task views with compact, editable table layouts

You can now use a compact layout option to organize fields in a table format within task views. Related information is easier to read and interact with, especially when you use multiple fields. You can define column and row headers, and fields remain editable based on access controls. The following figure is an example of a table layout for Budget and Schedule fields:

View grid data as cards for better readability

Users can now switch between grid and card layouts for data presentation. The card layout is especially useful on smaller screens and enhances readability for more consumable data. Admins can configure default and allowed layouts.

For more information, see Defining default layouts on Grid Views and Using Grid Views to find, open, and work with objects in the IBM OpenPages documentation.

Run calculations on selected objects from a grid view

Admins can now select specific objects from a grid view and run calculations on them. You save time and resources when you run targeted calculations.

Configure multiple relationships in object creation workflows

You can now configure multiple relationships in the workflow operations for creating objects. This enhancement simplifies complex object creation scenarios and improves data linkage.

For more information, see Defining a workflow action that creates objects in the IBM OpenPages documentation.

For more information about new features and other updates in Governance console, see New features in version 9.1.2 in the OpenPages documentation.

Create detached prompts in the UI

You can now create detached prompts directly in the user interface in both projects and deployment spaces. Previously, you could create detached prompts only by using notebooks. This enhancement simplifies detached prompt creation and improves accessibility for users.

Govern AI agents and tools

You can now find, add, and compare AI tools and agents in the governed agentic catalog. You can find tools that are provided by IBM, add endpoints for agents and tools that you deployed elsewhere, or add code for your custom agents and tools. You load and invoke agents and tools from the catalog in your solution by using the watsonx.governance Python SDK.

For more information, see Governed agentic catalog.

Evaluate agentic applications in Evaluation Studio

In the Evaluation Studio, you can now create an AI evaluation asset for AI experiment runs of an agentic AI application. You can then compare the results of the evaluations to identify the version of the application that performs best according to key criteria.

For more information, see Evaluation Studio for Agentic AI applications.

Govern Python functions to ensure transparency, traceability, and compliance for custom logic that is used in production workflows

You can now govern Python functions so that you can apply the same governance policies and metadata tracking to Python functions as you apply to models, datasets, and notebooks. This ensures transparency, traceability, and compliance for custom logic that is used in production workflows.

• You can now include Python functions in use cases, along with models and datasets, to provide context around which business problems they serve.
• You can include Python functions under different use case approaches—for example major or minor versions for evolving preprocessing logic.
• When tracked in a use case, the metadata of the Python function (such as owner, inputs and outputs, creation date, and associated compliance requirements) is captured.

Add user groups as collaborators on AI assets

When you add collaborators to an asset that is in an inventory, you can now select user groups as well as users. You can also assign roles to the users and groups.

For more information, see Adding collaborators to an AI use case.

For more information about these new features and other updates in Governance console, see New features in version 9.1.1 in the OpenPages documentation.

Compliance Accelerators for AI risk management

You can now subscribe to the Compliance Accelerators add-on to gain access to Policy Packs from Credo AI. The Policy Packs are a set of specialized, easy-to-use content bundles that provide clear, actionable insights aligned with global regulations and AI standards, including the EU AI Act, ISO/IEC 42001, and the NIST AI Risk Management Framework.

Default inventory replaces Platform assets catalog

Now you can use a default inventory to store artifacts including AI use cases, third-party models, attachments, and reports. The default inventory replaces any previous dependency on the Platform assets catalog or IBM® Knowledge Catalog for storing governance artifacts.

Multilingual Metrics Support in Generative AI Quality Monitor

You can now evaluate Generative AI outputs in multiple languages by using the Generative AI Quality monitor inwatsonx.governance. With this new feature, you can create both design-time and runtime evaluations using certain selected metrics for summarization, generation, extraction, question answering (QA), and retrieval-augmented generation (RAG) tasks. The languages include: English (en), Japanese (ja), German (de), French (fr), Spanish (es), Arabic (ar), Italian (it), Portuguese (pt), Korean (ko), and Danish (DA). For more information, see Evaluating generative AI output in multiple languages.

Extended Character Limit Support in Prompt Template Assets

You can now work with significantly larger text data in Prompt Template Assets (PTAs) by using a new column configuration option in the UI to increase the character limit value to more than 32k (excluding meta fields and label column in classification task type). This enhancement is ideal for use cases like large context for RAG, long-form summarization, transcript processing, and other large text scenarios in Generative AI workflow. For more information, see Evaluating prompt templates in projects and Evaluating prompt templates in spaces.

Drift v2 Batch Subscriptions support

You can now configure Drift v2 evaluations for batch subscriptions across all supported databases in IBM Cloud Pak for Data. This enhancement enables you to monitor model performance over time, ensuring consistent outcomes and data integrity in batch processing environments.

Evaluate global custom foundation models

You can now evaluate custom foundation models that are registered globally. With this enhancement, you can monitor and manage globally registered custom models that are accessed across multiple projects and deployment spaces. Inference operations use the Global Model ID, to ensure consistent model usage and streamlined governance across your enterprise environments.

For more information about these new features and other updates in Governance console, see New features in version 9.1.0 in the OpenPages documentation.