Release notes

Information about Workbench releases, improvements, and bug fixes
Tags:
Categories:

This document provides information about Verily Workbench product releases, improvements, and bug fixes. The most recent releases are listed first.

Last updated: March 30, 2026

2026

March 19: Support for GPUs on EC2 instances for AWS-backed cloud apps

Workbench now supports T4, A100, and H100 GPUs for JupyterLab, Visual Studio Code, and R Analysis apps on AWS-backed instances. For more details about configuring your AWS-backed cloud app, see Compute profile configuration options.

March 19: Support for NVIDIA cloud apps on AWS

Workbench users on AWS can now access cloud apps running NVIDIA NeMo and NVIDIA Parabricks and CUDA-X Data Science. These apps run inside containers that are built from a base NVIDIA image.

March 19: Support for a new pgweb cloud app on AWS

We have introduced a new platform client for PostgreSQL databases. This cloud app surfaces a simple yet powerful UI to explore Aurora databases on Verily Workbench.

March 19: AWS S3 bucket object versioning

Workbench customers on AWS can now request S3 bucket object versioning for controlled S3 storage folders. Users with reader/writer permissions can work with versioned S3 objects, including retrieving specific versions and deleting object versions. Reach out to Workbench support or your Workbench contact for more details.

February 26: AWS Aurora data warehouse resource

AWS Aurora PostGreSQL databases are now offered as a controlled resource in AWS data collections. See more about how to create an Aurora resource here, and how to query Aurora resources here.

February 12: Remove regional restrictions

Regional restrictions are now removed for workspaces within Workbench. Users may now create workspaces and resources in any available GCP region.

January 29: Egress override

Users can now request an egress threshold override in their exfiltration service. Researchers can then download files that abide by their data collection policies. See Exfiltration management for more details.

January 12: Metadata visibility for Discover-level users

Workbench users with Discover access can now view a limited set of the data collection's metadata, even if they're not members of the associated policy group. To learn more, see Data collection roles and access

January 10: Licensing tiers

Workbench has introduced a tiered system for user licenses. Existing users will occupy the Professional license tier with unrestricted access to all Workbench functionality. Users who self-register for Workbench will automatically receive the Standard license and are restricted to a subset of Workbench functionality. For more information on Workbench tiers, see Licensing tiers.

January 9: Instantly accessible data collections

Certain data collections in Workbench are now instantly accessible. Users can obtain immediate access after acknowledging a data use agreement. For more information, see Access data collections automatically

January 7: All of Us Researcher Workbench 2.0

The All of Us Researcher Workbench 2.0, powered by Verily Pre, is now available in Public Beta. Registered All of Us researchers may now create Verily Workbench workspaces connected to All of Us data.

Features such as Data Explorer, JupyterLab, and GitHub are now fully accessible to the All of Us community. Public Beta currently includes the All of Us Registered Tier data. Controlled Tier data will be available soon.

Learn more about how to use the new Researcher Workbench experience at Getting Started in new Researcher Workbench 2.0. If you would like to gain access to the All of Us data, follow the steps here.

2025

December 19: Create batch jobs

Workbench now allows users to execute workflows launched many times in parallel with different inputs. To learn more about running batch jobs in the Workbench web UI and CLI, see Create batch jobs.

December 19: AWS data collections support

Workbench now supports AWS data collections. This feature allows users to create and manage data collections with S3 resources. Resource addition will be strictly enforced, allowing only AWS resources (folders, S3 storage folders, external S3 buckets) into AWS data collections. To learn more about AWS data collections, see Considerations when creating a data collection.

December 18: Private custom apps

Workbench users can now configure custom applications using devcontainer.json files from a private Git repository. Previously, such files were required to be in public repositories. By using private repositories, users may use devcontainer.json configurations that contain proprietary or sensitive information.

See Publish your custom app to Git for more details.

November 24: Organization cloud audit logs

Cloud audit logs for BigQuery and Cloud Storage logs are now available to external customers. To ensure organizations only have access to their own data, each table is secured with row level security.

Logs are now accessible to organization administrators and Verily solutions engineers, enabling users to run queries joining the cloud audit logs and Workbench's service application data.

For more information, see Activity logs for organizations and Activity logs for data collections.

November 11: Workbench is now part of Verily Pre

Workbench is now part of Verily Pre, the AI-native platform for precision health.

Workbench is getting a new look as the user experience is unified across Workbench and all of Verily's solutions, including Refinery and Exchange. See Introducing Verily Pre for more details.

October 10: Data Explorer cohort creation from Marketplace to Workbench

Data Explorer now allows users to explore purchased data collections accessible through Workbench's Exchange page.

For more information, see Create cohorts from the Data Exchange.

October 7: Offer H200/B200 GPUs on VWB (GCP only)

Workbench applications can now attach NVIDIA H200 or B200 GPUs.

Access to these GPUs is limited and requires reservations. For more information, see Compute resource reservations for cloud apps.

October 7: NVIDIA applications

Workbench now offers two new NVIDIA cloud app types: NVIDIA NeMo and NVIDIA Parabricks and CUDA-X Data Science. Both types leverage GPU computing to support generative AI tasks (NeMo) and faster statistical, genomic and multiomics analyses (Parabricks and CUDA-X DS).

To learn more, see Use NVIDIA apps.

September 30: Verily Workbench Terms of Service (ToS) updates

The Verily Workbench Terms of Service have been updated to enable logged-in users to request read access to a data collection and have it automatically be granted. Other ToS updates include:

  • New trial users are on a time-limited (e.g., 90 days) trial/demo license. Verily reserves the right to delete these accounts at a future stage or modify/migrate/downgrade the account to align with our Standard Researcher license (once implemented). Any data brought by these accounts is subject to deletion when the license expires.
  • Verily reserves the right to remove the ability to access certain features within the product at a future date.

To learn more, see the Terms of Service.

September 30: Data Discovery becomes Data Exchange

Data Discovery has been renamed Data Exchange.

September 30: App constraint policy

A new app constraint policy limits the applications that can run within a Workbench workspace. When there's an app policy on a data collection, the list of available cloud apps will be limited to those defined by the policy.

See App policy for more information.

September 23: Enable org admins to use workspace AoD

Org admins can now use Access on Demand (AoD) into a workspace via the Workbench CLI. CLI instructions are available at wb workspace access.

The AoD access period for a workspace is one hour.

September 23: Network policy

Workbench has added a network policy that limits internet access for batch virtual machines (VMs), including Dataproc VMs and Google Batch VMs. When a data collection has a network policy turned on and is added to a workspace, all batch VMs in that workspace will be unable to send or receive network traffic.

See Network policy for more details.

September 9: Support for accessing external ECR repositories

Users can now create and reference external ECR repository resources, to enable AWS workspaces to access ECR repositories.

For more information, see Configure external ECR repository resources.

September 4: Application enhancements

Workbench has implemented cached application support across all of its GCP applications. This change drastically improves the user experience by reducing initial application startup time by up to 75 percent.

August 28: Workspace required form policy

Data collection owners can now enforce that Workbench users fill out a Required Form when adding their data collection to a workspace. In particular, a Use Statement must be completed when using data from the All of Us Research Program. This provides transparency about how data from various programs (e.g., All of Us) are being used, and enables data stewards to effectively review workspaces for policy compliance.

To learn more, see Required form policy.

August 28: GitHub integration design and refinement

Workflow definition files can now be read from linked GitHub repositories in a given workspace, allowing users to manage their WDL files with version control.

August 21: Discover data collections

The new Workbench Discover page lets researchers view and search across data collections that they have access to by name and by descriptive metadata tag. Users can click the Get access button on a given data collection to notify the data owner of their interest and obtain access.

See Find existing data collections for more details.

August 7: Workbench Terraform provider

Workbench now offers a Terraform provider to facilitate the management of resources like workspaces, data collections, and more through an Infrastructure-as-Code (IaC) approach. You can now:

  • Provision Workbench resources declaratively in code
  • Version-control your infrastructure
  • Ensure consistency with automated provisioning

This feature was designed to support the needs of Workbench users who want to provision workspaces and data collections at scale. This includes external customers who often want to manage Workbench resources the way they’re already managing related cloud resources.

For more information, see Get started with the Workbench Terraform provider.

July 10: Enable logout after inactivity

Workbench can now log out a user after a period of inactivity. The specific period of time allowed is set at the org level. For more details on how to set the period of inactivity, see Managing users.

July 2: Support for NVIDIA A100 & H100 GPUs

Workbench users can now add and edit NVIDIA A100 and H100 GPUs to their cloud applications through both the Workbench UI and CLI.

June 30: Migrate Workbench GCP workflow

Workbench’s GCP workflow has been migrated from the Google Cloud Life Sciences API (deprecated on July 8, 2025) to Google Cloud Batch API. Google Cloud Batch API supports all use cases previously covered by the Cloud Life Sciences API. The Workbench team has already adjusted permissions for new and existing projects to ensure seamless continuity.

For more information about workflows and Batch API, see Workflows overview.

June 16: Exfil prevention & egress monitoring

Workbench can now be configured to monitor egress traffic from all workspaces within a VPC-SC perimeter to the public internet. This service can be set up to take actions such as stopping user applications per pre-defined policies, as well as hitting an external webhook for notification purposes and additional customer-defined actions.

June 11: Allow editing compute environments after creation in UI

Users can now edit the machine type and disk size of Google Compute Engine apps via the Workbench UI. The CPU and memory fields are now configured by the machine type selection.

For more information, see Update compute options for an existing app.

June 5: Enable Vertex AI services

We are enabling the GCP Vertex AI APIs in Workbench workspaces backed by GCP projects. We will also grant permissions to workspace users to call those APIs.

For more information, see Access Vertex AI from your app.

April 22: Workbench groups 2.0

Workbench groups have been reconfigured. All newly created groups must now be linked to an organization. Legacy groups will continue to function as normal.

Additionally, new group roles include:

  • Admin
  • Member
  • Reader

An admin of a group must also be a member of the group to be able to access all resources that have that group as the group policy. The reader role allows a user or group to view that group's membership.

For more information about these changes, see Organization-scoped groups and legacy groups.

March 7: Support for self-service billing

Workbench now allows users to add their own Google Cloud Platform (GCP) billing information directly through the user interface.

Key updates include:

  • Google Account authentication: In the Profile section, users can authenticate with a Google Account that has access to a desired billing account.
  • Create pods with custom billing: Users can then navigate to the Pods section and create new pods using a billing account their Google Account has access to.

March 7: UI modifications for organization and pod management

Organizations and pods can now be directly managed via the Workbench UI. This will allow organization and pod admins to:

  • Add and remove users to organizations and pods
  • Change the roles of a given user in an organization or pod
  • Create pods
  • View and edit basic metadata on an organization and pod
  • View member lists and associated permissions for organizations and pods

For more information about these updates, see Set up billing with pods, Manage organizations, and Manage users.

January 21: Google Compute Engine JupyterLab apps

Workspace JupyterLab apps will now use a custom Google Compute Engine (GCE) configuration, replacing Vertex AI notebooks.

All new JupyterLab apps must use the new GCE configuration. To get started with a JupyterLab Compute Engine instance, see Create a new cloud app.

Have questions?

If you have any questions or need assistance, additional resources are available.

Last Modified: 26 March 2026