Access control

To use the generative AI features on Vertex AI, you must grant the appropriate IAM roles to principals, such as users, groups, and service accounts. You can grant a broad, predefined role or create a custom role with a specific set of permissions.

This page describes the different ways to grant access to generative AI features:

• Choosing a role: Learn about the different types of roles and how to select the best one for your needs.
• Predefined roles: Use broad, Google-managed roles for common use cases.
• Permissions for custom roles: Create custom roles by combining specific permissions for fine-grained access control.

Choosing a role

Vertex AI offers predefined roles for common use cases and the ability to create custom roles for more granular control. The following table compares these options to help you choose the right role for your principals.

Role	Description	Pros	Cons	Best for
Vertex AI Administrator (roles/aiplatform.admin)	Grants full access to all Vertex AI resources, including generative AI features.	Simple to manage; provides comprehensive permissions.	Violates the principle of least privilege; grants excessive permissions if a user only needs specific features.	Administrators who need to manage all aspects of Vertex AI.
Vertex AI User (roles/aiplatform.user)	Grants permissions to use Vertex AI resources, including making predictions and managing jobs.	Good balance for developers and data scientists who actively use the platform.	May still grant more permissions than necessary for specific, limited tasks.	Users who need to develop, train, and deploy models.
Custom Role	A role you create by combining specific permissions from the permissions list.	Follows the principle of least privilege; provides precise, fine-grained control.	Requires more effort to create and maintain.	Applications or users with specific, limited responsibilities, such as only making prediction calls.

Predefined roles

To give principals access to generative AI features on Vertex AI, you can grant one of the following predefined roles:

• Vertex AI Administrator (roles/aiplatform.admin)
• Vertex AI User (roles/aiplatform.user)

To learn more about Vertex AI IAM roles, see Vertex AI access control with IAM.

Permissions for custom roles

The following table maps generative AI operations to the permissions required for the operation. If you need fine-grained access control, you can refer to these mappings to create custom roles.

Operation	Permissions needed
Make prompt requests	aiplatform.endpoints.predict
Save, view, update, and delete prompts in Vertex AI Studio	aiplatform.datasets.create aiplatform.datasets.update aiplatform.datasets.delete aiplatform.datasets.list aiplatform.datasets.get
Model tuning	aiplatform.pipelineJobs.* aiplatform.customJobs.* aiplatform.datasets.export aiplatform.datasets.get aiplatform.models.upload aiplatform.models.get aiplatform.endpoints.create aiplatform.endpoints.get aiplatform.endpoints.deploy aiplatform.metadataStores.get storage.objects.create storage.objects.update storage.objects.get storage.objects.list

To learn more about Vertex AI IAM permissions, see IAM permissions.

What's next

• Learn how to enable Data Access audit logs.
• Learn more about access control with IAM.
• Learn how to tune a foundation model.
• Learn about responsible AI best practices and Vertex AI's safety filters.