What's new and changed in watsonx.governance

IBM watsonx™ Version 2.4.0

A new version of watsonx.governance was released in June 2026.

This release includes the following changes:

New features

This release of watsonx.governance includes the following features:

Evaluate detached prompt templates without creating deployments

Detached prompt templates are now evaluated without any dependency on watsonx.ai™. As a result, you now evaluate detached prompt templates directly in the deployment space, without creating deployments.

In a deployment space, you now start an evaluation from the Assets tab. Click the menu for a detached prompt template, and then click Evaluate.

You can no longer create deployments for detached prompt templates.

Your existing deployments are not impacted by these changes.

Watsonx.governance is now available on IBM Z and LinuxONE hardware

You can now install the watsonx.governance service on Red Hat OpenShift Container Platform on IBM Z and LinuxONE (s390x) hardware.

Create custom detectors in Guardrail Manager

You can now create custom guardrail detectors in Guardrail Manager. You create a custom guardrail detector by configuring the detector properties, input parameters, and the structure of the requests and responses for the detector.

Previously, you could create custom detectors only through the API.

Enhancements to Governance console

This release also includes enhancements to Governance console.

Solution enhancements

This release includes updates to help you manage AI tools:

• A new object type, AI Tool
• Views for the AI Tool object type
• A library business entity, Tool Library, to act as the parent for AI tools

This release includes a new library business entity, Agent Library, to act as the parent for AI agents.

Add favorite canvases to your dashboard

You can now mark canvases as favorites and display them in a new dashboard panel called Favorite Canvases. You can add favorite canvases in the user and admin dashboards for easier access and enhance productivity.

For more information, see Adding a Favorite Canvases panel.

Preview attachments

You can now view PDF, text, and image files directly in Governance console without downloading them. Files are rendered in an Governance console tab to ensure content is no longer accessible when the session expires.

The preview feature improves document accessibility and security by allowing users to preview attachments without leaving the Governance console interface or downloading files to their local system.

For more information, see Adding and managing all files (attachments).

For more information about new features and other updates in Governance console, see New features in version 9.2.0 in the OpenPages documentation.

New features from 5.3.1 patches

This release of watsonx.governance includes the following features that were introduced in IBM® Software Hub Version 5.3.1 patches:

Evaluate custom metrics for prompt templates by using LLM as a judge

You can now evaluate custom metrics for prompt templates and detached prompt templates by using the LLM as a judge (LLMaaJ) evaluator. With this capability, you can define domain-specific evaluation criteria beyond the built-in generative AI quality metrics. A large language model (LLM) acts as an evaluator (judge) to assess the quality of AI-generated responses based on customizable grading logic.

You can use the LLMaaJ evaluator in the user interface or by using the watsonx.governance Python SDK.

Compare custom metrics in Evaluation Studio

You can now use Evaluation Studio to compare custom metrics that you configured using a custom monitor. This capability supports only the endpoint integration for custom metrics.

Configure custom monitors for prompt templates from the UI

You can now configure custom monitors (metric groups) for prompt templates and detached prompt templates from the Projects and Spaces pages.

Update to the natural robustness metric

Watsonx.governance now uses the granite-4-h-small model to generate paraphrases for the natural robustness metric.

The granite-3-8b-instruct model that was used previously is deprecated.

For more information, see Natural robustness.

Migrate use cases to the latest version

You can now migrate AI use cases to V2 by using a command-line tool. Migrate use cases to ensure that your use cases are compatible with the latest features.

Use the usecase-migration command-line utility to do this migration.

For more information, see Migrating use cases with the usecase-migration utility

Updates

The following updates were introduced in this release:

Spanish and Japanese language support for RAG evaluation metrics

You can now evaluate generative AI models for retrieval-augmented generation (RAG) use cases in Spanish and Japanese. The following evaluation metrics now support Spanish and Japanese languages:

• Answer similarity
• Answer relevance
• Context relevance
• Faithfulness

Changes to creating detached prompt templates

When you create a detached prompt template, you must now provide a Foundation Model Name. In addition, the Foundation Model ID field is now mandatory.

Enhanced token count limits for Model Health Monitoring evaluations

Model Health Monitoring evaluations now accommodate significantly larger token counts for high-volume generative AI workloads. Individual token count fields (input_token_count and output_token_count) accept values up to 2,147,483,647, and aggregated total token count computations handle values up to 31 digits. This enhancement prevents numeric overflow issues when you run evaluations on workloads with very large token volumes and extended evaluation windows, which improves scalability and reliability for generative AI monitoring scenarios.

Customer-reported issues fixed in this release

For a list of customer-reported issues that were fixed in this release, see the Fix List for IBM Cloud Pak for Data on the IBM Support website.