ABAC on views entered beta in Azure Databricks, extending row-level policies beyond tables

The main development in October was the spread of attribute-based access control (ABAC) policies, which cover row-level security and column masking, to more kinds of objects. A community post on 8 October announced that ABAC on views was in beta in Azure Databricks [7]. A post on 11 October described ABAC as following the data to external engines and views [5]. On 7 October, a further post announced a beta for time travel on tables that use ABAC row-level security and column masking [8]. Together these posts suggest that policy coverage was widening beyond queries on current table data. The posts' titles are the only detail available here, so check the current documentation for exact scope, supported clouds and beta limits.

Practitioners also discussed how policies behave when they are wrong. A 14 October community thread covered fail-closed ABAC policies and early error detection [3]. The title points to concern about what happens when a policy cannot be evaluated, and about catching such errors sooner. Another 14 October thread asked about feasibility and security when using Lakehouse Federation to skip Bronze ingestion for near real-time pipelines [4]. A 9 October post on building a document knowledge base on Databricks also mentioned Row Filters [6].

On 26 October, a video showed how to govern AI agents with Unity Catalog and Unity Gateway. It covered access policies, user defined functions and role filters to control data and model interactions [2].

On 29 October, a v1.13.0 release was published. It added additive merging of tblproperties across config levels and opt-in invocation telemetry. It also included a breaking change for some view models, where view_update_via_alter alters Unity Catalog views in place and fails on Hive metastore views [1].