Databricks Explains OpenSharing Expansion Beyond Tables to AI Assets

Databricks published a November technical walkthrough of OpenSharing, the June 2026 evolution of Delta Sharing into an open, vendor-neutral protocol for sharing AI assets as well as tables. The vendor describes the change as additive rather than a replacement: existing shares, recipients and connectors continue to work, and new documented asset types are enabled through the same share, recipient and grant workflow. Databricks report.

What Changed in the Open Specification

According to Databricks, OpenSharing moves a broader asset model into the open specification itself. The protocol now defines a Share-to-Schema-to-Asset hierarchy that extends beyond Tables to Volumes, described as directories of files in any format, Agent Skills and AI models.

Databricks lists Agent and Page or Glossary types only as community proposals still under discussion, not finalized assets. Governance sits with the Linux Foundation AI & Data Foundation at sandbox stage, and the specification changes through a public proposal and review process.

Two Layers, and the Compatibility Question

The vendor draws a line between the open protocol and Databricks OpenSharing, its enterprise implementation integrated with Unity Catalog, Databricks Marketplace and Clean Rooms. Open source Unity Catalog provides a vendor-neutral catalog layer, which Databricks says lets a team pair it with an open sharing server and run governed sharing outside Databricks.

On compatibility, Databricks states that existing Delta Sharing clients and integrations remain compatible with OpenSharing, calling it an evolution of the protocol rather than a break from it. The earlier delta-io/delta-sharing repository continues to exist alongside the new OpenSharing-IO organization. The company says enabling OpenSharing on a Unity Catalog metastore is a configuration step, not a migration.

Documented Asset Types and Reach

The technical post describes support for Iceberg REST Catalog compatible clients, so providers can share to any Iceberg-compatible engine rather than only native Delta Sharing clients. It also covers on-premises and private cloud reach through the Databricks Storage Ecosystem, where storage vendors can run an OpenSharing server connected to Unity Catalog without first migrating data.

Zero-copy remains the model: the source says the sharing server issues temporary, scoped credentials and sits outside the data path. Databricks also lists capabilities built on top of OpenSharing, including SecureConnect at general availability, Global Distribution in private preview, Genie Agent Sharing in beta and Lakebase table and change data feed sharing in public preview.

What Remains Early or Proposed

The post distinguishes shipped work from commitments still pending. Foreign Iceberg table sharing was in public preview at the June announcement, with general availability described as coming soon, and email-based sharing was listed as coming soon at announcement time.

Beyond those, the project's public roadmap lists community proposals rather than finalized features, including scoped sharing of evaluation datasets and benchmarks, synthetic dataset sharing with provenance metadata and a proposed asset type for precomputed vector indexes.

What This Means for Practitioners

For teams already operating Delta Sharing, the practical implication is that existing tables, connectors and workflows remain in place while new asset types become optional additions. The same share, recipient and grant pattern applies, so adoption is a matter of configuration and choosing which assets to expose rather than rebuilding integrations, according to the vendor.

The stated value proposition is a single set of open APIs for discovery, authorization and access instead of bespoke pipelines per partner and asset type. Teams evaluating the claim should verify which capabilities carry general availability versus preview or beta status, and treat interoperability and governance benefits as vendor descriptions rather than demonstrated outcomes. Practically, that means testing against actual partner engines, including Iceberg clients, and reviewing the public specification and roadmap for the asset types relevant to their stack.