How is security handled in a serverless environment?

Summary

• Isolated network. The serverless compute plane runs in a separate network from the control plane, each workload runs on an isolated logical network with no public IP addresses, and firewall rules block inbound access and lateral movement between compute resources.
• Dedicated, wiped compute. Each node runs in a virtual machine assigned to a single customer, and nodes and attached resources are securely wiped before any reuse, with no cross-workspace reuse.
• Encryption everywhere. Traffic is encrypted in transit with TLS 1.2 or higher, and storage and attached disks are encrypted at rest with AES-256; customer-managed keys are available.
• No cross-tenant access. Workloads are isolated with no cross-workload communication, and data access uses short-lived, least-privilege tokens scoped to specific paths.
• Monitored and verified. Automatic security updates, security health monitoring, and independent third-party penetration testing back the platform's isolation controls.

How is security handled in a serverless environment?

With Databricks serverless compute, Databricks runs and secures the compute infrastructure for you. Security is built in through defense in depth: network isolation, dedicated and securely wiped compute, encryption in transit and at rest, strict tenant isolation, and continuous monitoring. For the full picture, see the Databricks Security and Trust Center.

How the Databricks Platform secures serverless compute

• Network isolation. The serverless compute plane operates in a network separate from the control plane, and all communication between them uses transport-level encryption. Each workload runs on an isolated logical network with no public IP addresses, and cloud security groups and firewall rules prevent inbound connections from external sources and block lateral movement between compute resources. See Serverless compute plane networking.
• Dedicated compute that is securely wiped. Each cluster node runs in a virtual machine assigned exclusively to a single customer, and nodes and their attached resources are securely wiped after use. Where a VM is reused within a single workspace, it is first wiped and reset to its original state, and there is no reuse across workspaces.
• Encryption in transit and at rest. All traffic between the user, control plane, compute plane, and cloud APIs is encrypted by default using TLS 1.2 or higher. Storage in the control plane and attached disks are encrypted by default with AES-256, including the internal storage bucket and remote result cache. You can optionally enable customer-managed keys for code, queries, results, secrets, and the storage bucket.
• No cross-tenant access. Workloads are isolated so no cross-workload communication is allowed, and each request is directed to a single workspace, with backend services isolating metadata and resources by customer and workspace. Access to data uses short-lived tokens with the minimum required permissions, scoped to specific paths for Unity Catalog customers.
• Layered defense in depth. Security is enforced across multiple layers at once: unprivileged, hardened containers; dedicated, encrypted disks; isolated VMs; and network boundaries between compute resources. A successful attack would need to defeat controls at the container, VM, disk, and network levels.
• Monitoring and independent verification. Databricks applies automatic security updates and provides security health monitoring so customers can assess their deployments against security best practices. The serverless isolation model has also been independently verified through third-party penetration testing that confirmed node isolation, secure wiping of resources, and prevention of lateral movement. Optional private connectivity, including private endpoints and storage firewall configurations, further restricts access to your cloud resources.

Getting started

• Review the Databricks Security and Trust Center for the platform's security model and certifications.
• Read Serverless compute plane networking to understand network isolation and connectivity options.
• Configure managed security features such as automatic updates, and enable customer-managed keys and private connectivity where required.

FAQs

Is serverless compute shared between customers?

Each cluster node runs in a virtual machine dedicated to a single customer, and resources are securely wiped before any reuse. There is no reuse of compute across workspaces and no cross-workload communication.

Is data encrypted in a serverless environment?

Yes. Traffic is encrypted in transit with TLS 1.2 or higher, and storage and attached disks are encrypted at rest with AES-256. Customer-managed keys can be enabled for code, queries, results, secrets, and the storage bucket.

Can I control network access to serverless compute?

Yes. Workloads run on isolated networks with no public IP addresses, and you can configure private connectivity such as private endpoints and storage firewall settings to further restrict access to your cloud resources.

How is serverless isolation verified?

Databricks applies automatic security updates and security health monitoring, and the serverless isolation model has been independently verified through third-party penetration testing of node isolation, secure wiping, and prevention of lateral movement.