How do I control access to serverless resources?

Summary

• Use serverless compute access control to govern who can run serverless workloads, through built-in compute objects in each workspace.
• Two default compute objects cover the main paths: Default Interactive Compute (notebooks and Databricks Connect) and Default Automated Compute (jobs and Lakeflow pipelines).
• Grant Can Use or Can Manage on those objects; by default all workspace users have Can Use, and admins have Can Manage.
• Restrict access in the workspace Compute → Serverless tab by editing permissions — remove the all-users group and add only the users, groups, or service principals you authorize.
• Control outbound network access with account-level network policies (serverless egress control), and govern the data serverless can read through Unity Catalog.

How do I control access to serverless resources?

On Databricks you control serverless resources on two axes: who can run serverless workloads and what those workloads can reach. Serverless compute access control governs the first through built-in compute objects; account-level network policies (serverless egress control) govern the second; and Unity Catalog governs the data any workload can access. See Manage serverless compute.

How the Databricks Platform controls access to serverless

• Serverless compute objects. Every workspace includes two default compute objects: Default Interactive Compute governs serverless for notebooks and Databricks Connect, and Default Automated Compute governs serverless for jobs and Lakeflow pipelines. By default all workspace users have Can Use on both, so existing workloads keep running. These default objects cannot be renamed or deleted.
• Two permission levels. Can Use lets a principal run workloads on that compute; Can Manage also lets them edit the object's permissions. Workspace admins have Can Manage by default and can grant it to others.
• Restrict who can use serverless. In the workspace sidebar, open Compute, go to the Serverless tab, and choose Edit permissions on the relevant compute object. Remove the group that includes all workspace users, then add only the specific users, groups, or service principals you want to authorize. If you revoke a user's interactive access, notebooks attached to serverless fail to connect and the serverless option no longer appears in their compute picker.
• Audit before you revoke. Before removing access to Default Automated Compute, review which jobs and pipelines the affected users own — a job already configured to run on serverless fails for any user who loses Can Use. You can query the system.billing.usage system table to see serverless usage by owner and by compute object.
• Manage permissions programmatically. You can set permissions on serverless compute objects with the Account Access Control API, which is useful for managing access at scale.
• Control outbound network access. Serverless egress control is configured through account-level network policies. A policy is either Full Access (unrestricted outbound) or Restricted Access (deny by default, allowing only Unity Catalog external locations and the FQDNs and storage locations you list). Policies support Enforced and dry run modes, and violations are logged in the outbound_network table in the system.access schema. A default network policy applies to all workspaces, and you can create additional policies for specific environments. See network policies.
• Govern the data, not just the compute. Unity Catalog applies fine-grained access control to the tables, volumes, and models any serverless workload reads or writes, so data governance stays consistent across compute types.

Getting started

• Read Manage serverless compute to set Can Use and Can Manage on the default compute objects.
• Open the Compute → Serverless tab and edit permissions to restrict serverless to specific users, groups, or service principals.
• Configure network policies to control outbound access from serverless workloads, and review the egress control overview.
• Govern the underlying data with Unity Catalog.

FAQs

What controls who can run serverless workloads?

Serverless compute access control, through two default compute objects: Default Interactive Compute (notebooks and Databricks Connect) and Default Automated Compute (jobs and Lakeflow pipelines). Grant Can Use or Can Manage on them.

How do I restrict serverless to specific users?

In the workspace Compute → Serverless tab, edit permissions on the compute object, remove the all-users group, and add only the users, groups, or service principals you want to authorize.

How do I limit what serverless workloads can reach on the network?

Use account-level network policies (serverless egress control). Restricted Access denies outbound connections by default and allows only the destinations you define, with violations logged to the system.access.outbound_network table.

Is serverless access control a spending cap?

No. It governs who can run serverless workloads. In a service disruption, access checks may fail open, so it should not be relied on as an absolute limit on billed amounts.