Databricks Unveils AI Gateway

Databricks is aggressively pushing into the agentic era with significant updates to its Unity Catalog, announced at the Data + AI Summit 2026. The platform, which now underpins governance for over 14,000 organizations, is evolving from a system of record to a real-time decision-maker for AI.

The core of these updates centers on providing control, context, and choice for AI deployments. Databricks is introducing the Unity Catalog AI Gateway, a new solution designed to govern the entire lifecycle of an AI agent's behavior. This moves beyond traditional access controls to manage how models, agents, and tools interact.

Control: Governing AI Actions

The Unity Catalog AI Gateway allows organizations to register and govern both Databricks-hosted and external models, MCP services, agents, and skills. It extends existing access controls, discovery, lineage, and auditing capabilities to these AI assets. Contextual Service Policies, currently in beta, enable administrators to define what AI agents can do at runtime, including allowing, denying, or requiring approval for specific actions like writing to sensitive folders. Databricks also highlighted AI Gateway budgets that now encompass external providers, offering a unified view of AI spend and hard spend caps.

Additionally, Databricks is introducing Governance Hub, a centralized command center for monitoring posture, identifying risks, and scaling governance operations across data, AI, cost, and performance. Attribute-Based Access Controls (ABAC) are being expanded with ABAC Grant Policies for models, automating permission grants based on attributes. Identity and Context Attributes will allow access rules to leverage live user properties and request context, moving beyond static group-based permissions.

Context: Empowering Agents with Business Meaning

To combat AI hallucination and ensure agents understand business context, Databricks is enhancing Unity Catalog Semantics. The new Glossary feature, coming soon, will allow for the definition of authoritative business concepts and terms, connecting them to underlying data and creating relationships. Domains, now in public preview, organize data and AI assets into business-aligned categories, providing scoped context for agents.

Metrics in Unity Catalog are also seeing significant expansion, enabling the definition of reusable business KPIs like revenue and churn. These metrics will support richer semantic modeling, including multi-fact relationships and level-of-detail calculations. Agentic and UI-driven authoring will lower the barrier to creating well-defined semantics, with Materialization offering faster query performance. External lineage is now generally available, extending lineage tracking beyond Databricks to encompass upstream source systems and downstream BI reports, providing agents with a more complete data flow picture.

Choice: Open Infrastructure for AI

Databricks is emphasizing flexibility with cross-cloud and cross-region addressability for Unity Catalog. A new four-level namespace will provide a single address for every asset across an organization's entire Databricks footprint, enabling unified discovery and consistent policy enforcement regardless of cloud or region. This move aims to prevent vendor lock-in and allow workloads to run where capacity exists.

These Unity Catalog new features 2026 are set to redefine how enterprises manage and deploy AI at scale, integrating AI governance directly into the fabric of data management. The updates signal a strong push by Databricks to lead in the emerging agentic era, as seen in related advancements like Databricks Bolsters AI Governance and Azure Databricks embraces agentic era.