Role-Based and Project-Based Access Controls for watsonx Models

There needs to be a more granular access control mechanism within watsonx that allows administrators to manage access at the individual model and agent level. Organizations should be able to approve, restrict, or revoke access to specific models based on internal governance policies, contractual agreements, licensing terms, risk reviews, and approved use cases.

Currently, users may have visibility into or access to models that are not approved for their company, business unit, project, or use case. This creates a governance gap because some models may require separate service agreements with third-party providers, may not be approved for certain types of data, or may only be authorized for experimentation rather than production use.

The requested capability should allow platform administrators to define which users, groups, roles, projects, or workspaces can access each available model. If a model is not approved or covered by the appropriate service agreement, users should be prevented from selecting, invoking, deploying, or integrating that model through any watsonx interface, including the UI, APIs, SDKs, CLI tools, notebooks, and agentic AI workflows.

For agentic AI use cases, administrators should also be able to control which models and tools an agent can use. This would ensure that agents cannot be configured with or dynamically invoke models that are not approved for the user, project, or organization. Access restrictions should be consistently enforced across both direct model usage and indirect usage through agents, applications, or automated workflows.

This functionality is needed to support enterprise AI governance, contractual compliance, risk management, and responsible model adoption. By enabling model-level and agent-level access controls, watsonx would allow organizations to ensure that only approved models are available for authorized users and use cases.