What is the Databricks security architecture on Azure?

Summary

• On Azure, the Databricks Platform separates a Databricks-managed control plane (workspace application, notebooks, configuration, and cluster management) from a compute plane where your data is processed, so permissions stay simple and data is not duplicated.
• Classic compute runs Databricks Runtime (Apache Spark) clusters inside your own Azure subscription; serverless compute runs in a Databricks-managed network separate from the control plane, with all cross-plane traffic encrypted at the transport level.
• Secure Cluster Connectivity eliminates public IPs on cluster nodes and is on by default: the compute plane has no open inbound ports and connects outbound only to the control plane.
• Private-by-default networking uses VNet injection, front-end and back-end Azure Private Link, network security groups, user-defined routes, and deny-by-default serverless egress control to keep traffic on trusted paths.
• Unity Catalog governs data and AI with fine-grained access control, lineage, and audit; access is authenticated with Microsoft Entra ID SSO, MFA, and SCIM; data is encrypted in transit and at rest with support for customer-managed keys.

What does the Databricks security architecture look like on Azure?

Azure Databricks is built on a defense-in-depth model spanning identity, network, compute, data, and governance, structured around two components: a Databricks-managed control plane that runs the workspace application, notebooks, configuration, and cluster management, and a compute plane where your data is processed. Separating data processing from the management layer keeps permissions simple, avoids data duplication, and reduces risk. Around that core, Azure Databricks layers customer-managed networking, Microsoft Entra ID identity, encryption, and unified governance through Unity Catalog so consistent controls apply across analytics, AI, and applications.

Why the Databricks Platform is secure on Azure

• Control plane and compute plane separation. The control plane hosts the workspace application, notebooks, configuration, and cluster management. With classic compute, Databricks Runtime (Apache Spark) clusters run inside your own Azure subscription. With serverless compute, Databricks manages the compute-plane networking in a network separate from the control plane, and all communication between the planes uses transport-level encryption.
• VNet injection for network control. For classic workspaces, you can deploy Azure Databricks into your own virtual network using VNet injection, giving you control over IP address ranges, routing, and network security groups. VNet injection is also required for classic Private Link.
• Front-end and back-end Azure Private Link. Private Link provides private network connectivity between your virtual network and the Azure Databricks control plane. Front-end Private Link enables private connectivity for user access and browser-based single sign-on, while back-end Private Link keeps REST API and Secure Cluster Connectivity relay traffic between clusters and the control plane off the public internet. For serverless, a dedicated Private Link connection can link serverless SQL warehouses to your storage, with a private endpoint dedicated to your account and reachable only from authorized workspaces.
• Secure Cluster Connectivity (SCC). SCC eliminates public IP addresses on cluster nodes and is enabled by default with no additional configuration. The compute plane has no open inbound ports, clusters run in a private network in your cloud account, and all connections to the control plane are outbound only.
• Network security groups and egress controls. Azure Databricks supports user-defined routes, firewall rules, and network security groups. For egress control, you can deploy a firewall (or a network virtual appliance) in a hub virtual network, define the destinations that traffic is allowed to reach, and route non-local traffic from Databricks subnets through it. For serverless, network connectivity configurations govern egress and establish private connections to storage, and serverless egress control enforces a restrictive policy that permits outbound traffic only to required destinations.
• Unified governance with Unity Catalog. Unity Catalog is a single governance solution for data and AI assets — including files, tables, and machine learning models — providing fine-grained access controls, centralized policy management, auditing, and lineage tracking. It issues time-bound, down-scoped credentials that are accessible only to trusted code by default.
• Microsoft Entra ID identity. User access is managed with single sign-on and multi-factor authentication via Microsoft Entra ID, and conditional access can further restrict authenticated access. SCIM synchronizes users and groups from your identity provider so access is provisioned and de-provisioned automatically.
• Encryption in transit and at rest. All communication between the control and compute planes uses transport-level encryption. Data is encrypted at rest, and customers can supply customer-managed keys for the control plane and workspace storage.
• Independent testing and assurance. Azure Databricks is tested with annual third-party penetration tests and a public bug bounty program, and is designed with at least two strong security boundaries at every point of system or user interaction.

Getting started

• Start with the Databricks security overview to see how identity, network, data, and governance controls fit together.
• Learn how to harden ingress and egress in data exfiltration protection with Azure Databricks.
• Follow the security best practices for the Databricks Data Intelligence Platform.
• Deploy secure defaults with the open-source Databricks Security Reference Architecture (SRA) Terraform templates, which include Azure components.
• Review current certifications at the Databricks Security and Trust Center and the Databricks Security Addendum.

FAQs

How does Azure Databricks separate management from data processing?

The Databricks-managed control plane runs the workspace application, notebooks, configuration, and cluster management. The compute plane processes your data — inside your own Azure subscription for classic compute, or in a Databricks-managed network for serverless. All traffic between the planes is encrypted in transit.

Does Secure Cluster Connectivity require configuration?

No. Secure Cluster Connectivity is enabled by default with no additional setup. It removes public IPs from cluster nodes, and the compute plane accepts no inbound connections while connecting outbound only to the control plane.

Can I keep Azure Databricks traffic off the public internet?

Yes. VNet injection, front-end and back-end Azure Private Link, network security groups, user-defined routes, and serverless network connectivity configurations let you keep user, cluster, and storage traffic on private, allow-listed paths.

How are users authenticated on Azure Databricks?

Access uses single sign-on with multi-factor authentication via Microsoft Entra ID, with optional conditional access. SCIM synchronizes users and groups so access is provisioned and de-provisioned automatically.

What compliance and assurance does Azure Databricks provide?

The platform undergoes annual third-party penetration tests and runs a public bug bounty program. See the Databricks Security and Trust Center and Security Addendum for current certifications and compliance details.