What’s new in Databricks - August 2026

Databricks shipped the following major GA launches in August 2026 :

• Unity AI Gateway is GA: Databricks’ governance layer for enterprise AI, covering model access, MCP governance, and cost observability.

• RBAC (Role-Based Access Control) is GA: lets Databricks users assume scoped roles instead of using accumulated permissions.

• Secrets in Unity Catalog is GA: secrets become governed, three-level-namespace securable objects.

• Serverless Compute Access Control is GA: admins can govern who runs serverless workloads in Databricks.

• Lakebase Postgres APIs are GA, alongside LTAP Direct Writes for faster synced-table loads.

• The Agent mode API for Genie Agents is GA and Full page Genie Code is GA.

🛠️Data Engineering

Databricks Spark Declarative Improvements

• You can refresh Materialized Views and Streaming Tables even when the pipeline owner or run as identity is subject to an ABAC.

• Materialized views and Streaming tables refreshes in Databricks SQL inherit custom tags from the enclosing SQL warehouse environment. The tags propagate to billing surfaces enabling proper cost attribution of materialized view and streaming table updates back to the source SQL warehouse

• The REPLACE USING flow simplifies ingestion when dealing with partial snapshots instead of traditional Change Data Capture (CDC) feeds.

  • Key-Based Group Replacement: Specify one or more key columns (such as region_id or snapshot_date). When new data arrives, the system completely overwrites all existing rows matching those keys while leaving the rest of the dataset untouched.

  • Out-of-Order Handling (SEQUENCE BY): Define a sequencing column (like a timestamp or version number). The engine automatically resolves late-arriving or out-of-order batches by ensuring the highest sequence value always overwrites older data.

  • Simplified Pipeline Logic: Eliminates the need to construct complex merge statements or maintain explicit CDC event types for bulk dataset updates.

• Databricks supports reading a Change Data Feed (CDF) directly from Materialized Views created via Lakeflow pipelines or Databricks SQL. This enables real-time tracking of row-level updates, insertions, and deletions within the view. Row tracking should be enabled => delta.enableRowTracking

• You can orchestrate continuous Lakeflow pipelines directly through continuous jobs in Databricks.

• Overview Job-level performance modes apply to Materialized View and Streaming Table refreshes triggered by dbt tasks. Previously locked to Performance-optimized mode, dbt-driven refreshes can now run in Standard mode to reduce compute costs.

  Environment variables for Serverless Jobs is Beta

You can easily inject deployment-specific configurations into your application code. Simply define environment variable groups at the job level and select the appropriate entry for each task. Configure environment variables for serverless jobs

New Lakeflow Connectors are available

• Google Drive Connector is GA

• NetSuite Connector is GA

• PagerDuty Connector is GA

• SharePoint Connector is GA

• Notion Connector is Beta

• Amplitude Connector is Beta

• Glean Connector is Beta

• Google Search Console Connector is Beta

• Gmail Connector is Beta

• SendGrid Connector is Beta

• Workiva Connector is Beta

• Zerobus Ingest into Streaming Tables is GA

🗄️Lakebase Postgres

• Lakebase Postgres APIs are GA

• Genie Code is available in the Lakebase Project UI

• LTAP Direct Writes accelerates initial data loads and full refreshes when creating or updating synced tables in Lakebase. By bypassing the Postgres compute endpoint, it streams Unity Catalog data straight to the storage layer, reducing load times and preventing query performance degradation.

🏢Data Warehousing

The Databricks Connector for Google Sheets lets you write data from Google Sheets back to a Unity Catalog table. You can create a new table or overwrite an existing table without leaving Google Sheets. Write data back to Databricks from Google Sheets.

🤖Databricks GenAI & Machine Learning Capabilities

• The ai_search() SQL function executes natural-language searches across indexed knowledge sources. It automates query expansion, cross-source retrieval, deduplication, and relevance reranking to return the most pertinent documents. Ai_search function.

• The ai_prep_search() SQL function accepts plain text and markdown STRING content in addition to the structured VARIANT output of ai_parse_document. You can chunk and prepare text that you already have as a string for AI Search indexing and RAG pipelines, without first running it through document parsing. function.

• You can build custom trace views in the Databricks MLflow trace explorer. Describe the layout you want in plain language and Genie generates a reusable view that surfaces the trace fields, metrics and feedback controls most relevant to your review workflow. Build a custom trace view.

• The unified trace table captures every request and response across your Unity AI Gateway services into one Unity Catalog table in OpenTelemetry format, giving you a single place to monitor, debug, secure and audit all AI activity. Monitor all AI activity using the unified trace table.

🖥️ Databricks Workspace Platform and Compute

Serverless Compute Access Control is GA

Databricks has deployed serverless compute access control, giving workspace administrators centralized authority over who can execute serverless workloads. The administrators can control access across two distinct built-in serverless compute profiles:

• Default Interactive Compute: Governs execution permissions for notebooks and Databricks Connect.

• Default Automated Compute: Governs execution permissions for automated jobs and Spark Declarative Pipelines on Lakeflow.

What’s the immediate Impact?

• Zero Downtime or Disruption: All workspace users are automatically granted Can Use permissions on both objects by default, ensuring existing workloads run uninterrupted.

• Granular Governance: Admins can restrict or customize these default permissions to enforce tighter security policies across interactive and automated serverless tasks.

Organize your work with spaces

Spaces can be created manually or automatically by opening a pipeline or Declarative Automation Bundles. You can also open a Git folder as a space from the workspace browser. Spaces are private to each user account.

When you’re in a space:

• Your open tabs are saved with the space.

• The sidebar file tree opens at the space’s folder.

• The workspace search bar defaults to searching in the folder.

Excel Add-in

The Databricks Excel Add-in has been updated to make filtering importied datasets smoother and more intuitive and closer ot native excel workflows by supporting cascading filters and case-insensitive string matching. LIKE, NOT LIKE, IN, NOT IN, and ILIKE filters are no longer available in the filters list. Filter imported data.

Search and replace text across multiple files simultaneously.

You can search and replace text across all files at once within Databricks Git folders and Declarative Automation Bundles projects. Search and replace across files.

Based environments On classic compute

You can use base environments to manage Python dependencies on classic compute. You can use Databricks-provided environments or custom workspace-level environments.Manage dependencies using environments.

Session restore for Serverless Jobs is Beta

Restore the Python variables and Spark session from a serverless compute job run into a new interactive notebook, so you can debug a failure or explore results without rerunning the job. Databricks automatically captures a state snapshot when an eligible run fails, runs 30 minutes or longer, or is cancelled and keeps it for 7 days after the run completes. Session restore for serverless jobs.

🛡️Unity Catalog and OpenSharing

OpenSharing Feature Rollout

• Sharing Delta Tables with Deletion Vectors and Column mapping is GA.

• Sharing Delta Tables with Iceberg reads enabled is GA.

• Sharing Foreign Delta Tables with OpenSharing is Public Preview.

• Sharing tables backed by default storage is GA.

• Sharing data behind a firewall with SecureConnect is GA.

• Sharing managed Iceberg tables is GA.

Unity Gateway is GA

Unity AI Gateway is the Databricks governance solution for enterprise AI, part of Unity Catalog. The GA capabilities include:

• Unity Catalog governance: permissions and access control for models, model services, external model providers, MCP services, functions, and connections.

• Model access and provider integration: Databricks-hosted models, external providers such as OpenAI and Anthropic, and governed model-provider services.

• Traffic management: central routing, rate limits, traffic splitting, and fallbacks.

• MCP governance: registration, permissions, tool filtering, managed/custom MCP services, and on-behalf-of execution.

• Usage and cost observability: usage tracking, cost attribution, dashboards, budgets, and spend caps.

• Coding-agent integrations: governed access through supported coding-agent integrations such as Claude Code, Cursor, Gemini CLI, Codex, and others.

Those capabilities remain in Beta:

• Service policies / contextual policies including built-in and custom guardrails for PII, prompt injection, unsafe content, jailbreaks, and policy violations.

• Agent services registering and governing external agents as Unity Catalog securables.

• Smart Routing: automatically selecting models

• Skills: governed, reusable agent capabilities in the Unity Catalog inventory.

• Unified trace table / AI activity tracing centralized tracing of model and MCP interactions.

MANAGE Privilege Usage Requirements Removed

The MANAGE privilege no longer requires a usage privilege (USE CATALOG or USE SCHEMA) on the object where MANAGE is granted. This change makes MANAGE behave more consistently with object ownership.

Secrets in Unity Catalog is GA

You can store, govern, and access secrets as securable objects in Unity Catalog. A Unity Catalog secret uses the three-level namespace (catalog.schema.secret) and is governed by Unity Catalog privileges, so you can apply the same access controls and auditing that you use for other data assets.Secrets in Unity Catalog.

Databricks managed MCP connectors are integrated with Unity Gateway

All Databricks-managed MCP connectors for Genie One and Genie Code have migrated to the Unity AI Gateway.

Tags Update

Tag Automations dynamically assign or remove governed tags on Unity Catalog tables and volumes based on user-defined rules, keeping metadata accurate as underlying assets evolve.

The Tags page provides a centralized view of how tags are applied across your account. It summarizes governed tag usage, shows recent tag assignments and their sources, and surfaces recommendations to fix invalid tag values and add tags to important assets that are missing them.

Observability & System Tables Roundup

• System.alert schema contains the alerts table which record the configuration of every alerts and alert_evaluation_history table record one row per alert evaluation.

• Query text is masked by default only Account admins and members of the databricks_pii_access account group can read unmasked query text.

• Pipeline_events records Lakeflow pipelines event log entries capturing lifecycle transition, flow progress, data quality metrics, error and operational data across all pipelines. Pipeline events system table reference

ABAC Expansion

• Attribute-Based Access Control (ABAC) GRANT policies dynamically assign Unity Catalog privileges to securable assets whenever their governed tags satisfy specific condition rules. ABAC GRANT policies

• ABAC GRANT policy support has expanded to cover additional AI and serving infrastructure:

  • Models

  • Model Serving Endpoints

  • External / Provider Models

  • Model Context Protocol (MCP) Services

  • Agent Services

• Databricks syncs up to nine Identity Provider attributes (such as job title, department, or cost center) (Beta)directly onto account user profiles for use in Unity Catalog Attribute-Based Access Control column-masking policies.

• You can write Unity Catalog ABAC column mask policy conditions that target users by attributes synced from your identity provider in addition to group membership( Beta) has_identity_attribute_value & has_identity_attribute_tag_match functions are supported in the WHEN clause of column mask policies.

• Context Attributes for ABAC Policies (Beta) allow Unity Catalog row filters and column masks to enforce access rules based on request metadata such as caller identity and execution mode rather than user identity alone.

• Key Context Parameters & Functions

  • Supported Functions: Evaluated in WHEN clauses using has_context_attribute and has_context_attribute_value.

  • Request Metadata: Evaluates variables like request.client_id (the calling OAuth app) and request.is_on_behalf_of (user-delegated calls).

New Permissions in Unity Catalog is Beta

Unity Catalog introduced granular INSERT, UPDATE, and DELETE permissions.These provide a least-privilege alternative to the broad MODIFY permission, allowing you to grant specific data manipulation rights without exposing table structure or metadata modifications. Fine-grained DML privileges.

• INSERT: Permits adding new rows to a table.

• UPDATE: Permits modifying existing rows in a table.

• DELETE: Permits removing existing rows from a table.

Unity Catalog Skills is Beta

You can publish, govern, and share agent skills with Unity Catalog. A skill is a first-class Unity Catalog securable (catalog.schema.skill) that follows the SKILL.md specification, so you register it once and control who can discover, use and update it with the same permissions, tags, and audit that protect your tables.

Coding agents download published skills or load them live over MCP What are Unity Catalog Skills.

RBAC is GA

RBAC lets users assume a role in Databricks. When a user assumes a role, only that role’s permissions apply, rather than the user’s accumulated permissions. This enables exclusive access. Users must actively assume a role to access sensitive data, preventing them from accessing it under their own identity and from mixing data across use cases, clinical trials, projects, or clients. Role-based access control (RBAC).

Genie Ontology

Pages serve as standard, governed definitions for business terms, acronyms, and KPIs. Managed as part of Unity Catalog semantics, they make up the curated, human-modeled layer of the Genie Ontology. Pages.

📝Lakeflow Designer

📝Genie Agents

• Programmatically extracts charts for downstream report, enabling automated scripts to post weekly visualizations directly to executive in Slack. Retrieve visualization results.

• Content search for attached volumes (Beta): Enable content search for attached volumes.

• Analyze files in volumes without Unity AI Gateway (Beta): Analyze files in volumes with a Genie Agent.

• Prevent premature execution by submitting files and instructions together. Upload a file to a Genie Agent.

• The conversation list API returns conversations from the public API, Genie One, and embedded Genie, so the conversations it returns match the history shown on the monitoring page. Use the Genie Agents API.

• You can provide creation/update metadata for programmatic tracking, allowing developer portals to display only recently modified Genie Agents.

• Markdown tables in Agent mode APIs (Beta): In the Agent mode APIs, table visualizations now return as markdown instead of table visualization attachments, which improves readability. Parse the report.

• Longer agent descriptions: Speeds up agent discovery by displaying full scope details in list views, helping analysts immediately pick the right tool for specific datasets like EMEA supply chain.

• Comment on answers from Genie One.

• The list conversations APIs include a field that indicates whether a conversation was sent in Agent mode or Chat mode.

• Download CSV in embedded agents.

• The Agent mode API for Genie Agents is GA

• Agent authors can attach both scalar and table-valued Unity Catalog functions as tools for a Genie Agent.

📲Genie One

• Scheduled task consistency: Scheduled tasks look up past runs to keep run-to-run outputs consistent. Scheduled tasks.

• Export documents to PDF: Simplifies sharing with non platform stakeholders, allowing analysts to convert insights into PDF decks. Documents in Genie One.

• Ontology snippets is available to all customers (Public Preview): Genie Ontology.

• Workspace instructions (GA): Enforces global business context and compliance rules across all workspace chats, ensuring every user prompt adheres to standardized enterprise guidelines. Add workspace instructions for chat.

• Upload images and Word documents: Genie One supports uploading images (PNG, JPG, and JPEG) and Word (.docx) files to a conversation, in addition to CSV and Excel files. Upload a file.

• MCP writes (Beta): Unlocks automated operational workflows, enabling connected Model Context Protocol tools to update CRM records or trigger external jobs directly from chat outputs. Connect to external tools and sources.

• PDF attachments for scheduled tasks: Scheduled tasks can now send their results as a PDF email attachment. Scheduled tasks.

• Upload PDF files to a conversation (Beta): You can now upload PDF files to a Genie One conversation alongside CSV and Excel files. Genie One parses PDFs asynchronously. Upload a file.

• Mention workspace users in document comments: accelerates collaborative review inside generated reports, letting creators tag subject matter experts (@user) for quick feedback on specific findings.

• Genie Ontology enabled by default (Public Preview): Delivers instant business-aware context out of the box, ensuring user queries automatically yield enterprise-accurate definitions without manual admin setups.Genie Ontology

• Genie mobile (Public Preview): Eliminates authentication friction for multi-tenant accounts, allowing leads to toggle between development and production environments on mobile in one tap and Maintains strict zero-trust security without blocking access, enabling engineers connected via corporate VPNs to securely sign into mobile workspaces behind IP access lists.

• Open Unity Catalog tables in Genie: You can now open Unity Catalog tables directly in Genie without navigating to Catalog Explorer Open tables.

  • Tables in chat: Opening a table from chat keeps you in the conversation. View overview details, columns, and sample data next to your chat.

  • Tables in Discover: Opening a table from Discover shows a full-page catalog-style view optimized for Genie.

  • Ask Genie: From a table in Discover, click Ask Genie to start a chat about that table.

  • Request access: If you only have browse access, you can request access from the table view.

• Pin chats: You can now pin a chat in the Genie One sidebar to keep it at the top for quick access. Manage chats in the sidebar.

• Rename chats: You can now rename a chat in the Genie One sidebar. Manage chats in the sidebar.

• Visualizations in scheduled task emails: Scheduled task email outputs can now include visualizations.Scheduled tasks.

• Genie One desktop app (Beta): The Genie One desktop app, a native macOS app, is now available. Chat with your data and work with Genie without a browser. See Use Genie One on desktop.

• Create scheduled tasks in the Genie mobile app (Public Preview): You can now create scheduled tasks directly from the Genie mobile app Scheduled tasks.

📊AIBI Dashboard

• Import from Power BI and Tableau (GA): You can import reports and data models from Power BI and Tableau with Genie Code (/importBI). Import BI files using Genie Code.

• Larger Excel downloads: You can download up to 1 million rows from a dashboard widget to Excel (.xlsx), up from 100,000. Download results.

• Multi-select and move widgets: Dashboard authors can select multiple widgets and move them together on the canvas by holding Cmd or Ctrl while clicking. Keyboard shortcuts.

• Hide the sparkline and date labels on counters: You can hide the sparkline and date labels on counter widgets. Hide date labels and the sparkline.

• Duplicate subscriber prevention in the API: Dashboard subscription APIs prevent duplicate recipients from being added.

• Unsubscribe removes duplicate subscriptions: Unsubscribing from a schedule by email or from the dashboard removes all duplicate subscriptions for that recipient. Manage scheduled dashboard updates and subscriptions.

• Local metric views (GA): You can define fields, measures, and join relationships directly in an AI/BI dashboard with a visual interface, without first publishing them to Unity Catalog. Local metric views.

• Low-code UI for Unity Catalog metric views in dashboards: When you add a Unity Catalog metric view to a dashboard, it now uses the same low-code UI as local metric views, replacing the legacy UI. Extend a Unity Catalog metric view.

• Remove duplicate subscribers: Dashboard authors can remove duplicate subscribers from a dashboard schedule. Manage scheduled dashboard updates and subscriptions.

• Rich text in widget titles and descriptions.

• Dashboard authors can hide the column header in a pivot table.

• Dashboard authors can see the run history for a schedule, including the run ID, start and end times and status for each of the five most recent runs. View schedule run history.

• Genie Code for dashboard authoring:

  • You can undo and redo edits made by Genie Code. Use Genie Code for dashboard authoring.

  • Genie Code can optionally materialize datasets.

  • You can manage dashboard email schedules directly from Genie Code.

  • Genie Code can set a schedule’s tabular attachment, so a scheduled email includes widget data. Email attachments.

  • Genie Code places widgets in appropriate locations instead of leaving large gaps in the dashboard layout. Dashboard layout and organization.

🔮Genie Code

• You can now set the effort level for Genie Code to balance response quality and cost. Auto (default) gives you the highest quality responses. For simpler tasks, select Low to use a faster, lower-cost model. Select a level of effort.

• Full page Genie Code is GA. Full page Genie Code.

• Genie Code can search the public web to answer questions that need current information Search the web.

• You can now upload files to Genie Code to use as context for the current chat. See Attach files.

• Genie Code is available for AI Runtime to develop and troubleshoot deep learning workloads

• Genie Code is available as a Lakeflow Jobs Task. It can read upstream task outputs and call tools as needed. Genie Code task for jobs.

Originally published on Databricks Community