5 New Developments in Databricks and How Securiti Customers Benefit
Concerns over the risk of leaking sensitive data are currently the number one blocker for organizations dedicated to deploying AI solutions now.
At this year's Data and AI Summit, Databricks introduced several capabilities that, with the integrated capabilities of Securiti address this issue by helping the largest companies in the world streamline unstructured data ingest and accelerate deployments of permission-aware AI applications. Organizations can do this by using Securiti’s Gencore Sync to find unstructured data across hundreds of systems, dig into those unstructured sources at the element level, obfuscate and sanitize all sensitive data, and then sync directly to Delta tables. As a result, Databricks users can gain contextual insights across environments, including access and policies enforced through Unity Catalog.
For companies already using Securiti's Data+AI Command Center, here are the key announcements that create new possibilities for maintaining strict permission boundaries while expanding AI use cases.
Two Core Problems
An estimated 80-90% of enterprise data lives in unstructured formats—documents, emails, presentations, videos. When this data feeds AI applications, organizations face two fundamental challenges:
- As data continuously pours in, how do you know what’s buried in those files to ensure you don’t inadvertently expose sensitive data?
- How do you maintain the same access controls that exist in source systems when data moves through AI pipelines?
Traditional approaches to classification of unstructured data typically occur after data ingestion, creating risk and latency. Manual approaches do not scale well and often rely on sampling strategies or metadata-only, leaving unknowns in the data. An example could be a marketing team processing customer reviews could unknowingly ingest and process PII that could be later output by the AI.
Securiti addresses this by deep scanning existing data across environments as well as during ingestion. This provides accurate classification and labeling of all sensitive data, to be continuously synched with repositories for training or reference data, reducing latency and risk.
Take for example, sensitive HR data- A user who has access to AI should be able to query files related to company policies and programs, but should NOT be able to obtain information about employee salary unless they are specifically entitled to the files that contain that info. Most systems unfortunately can't enforce this consistently.
Securiti addresses this by extracting file level permissions from source systems, and persisting them to be applied through the entire pipeline. When unstructured data gets sanitized and synced to Delta tables, the permission graph travels with it. AI applications built on this data remain "permission aware"—users only see outputs based on data they could originally access.
What Changed at DAIS 2025
Agent Bricks: No-Code AI Agents
What’s new: Databricks introduced Agent Bricks for building domain-specific AI agents. Agent Bricks allows a user to choose an agent template, describe a problem in natural language and then sit back as the tool designs the agent, suggests benchmarks, synthesizes data, fine tunes and optimizes for performance/cost tradeoffs. Agent Bricks leverages Databricks Lakehouse including Unity Catalog to provide governance for production-grade Agents
Why it matters: Agent Bricks seeks to solve the “too many knobs” problem blocking agent development. By offering automation in many of the technical aspects of designing an agentic system, Databricks lowers barriers and shifts the focus of agentic development from technical details to business criteria.
How Securiti and Databricks are better together: Securiti ensures agents only work with data users are entitled to see. When an agent processes a request, Securiti is able to apply access context from the underlying data. Fresh data can be redacted, anonymized, dynamically masked or otherwise sanitized upon ingestion to responsibly accelerate the use of agents with safe data.
Unity Catalog Evolution
What’s new: Unity Catalog now supports better automated access controls and can ingest more detailed data classification with attribute-based access control (ABAC), automated data classification that uses language models to infer from metadata, and tools to support unified governance across storage formats.
Why it matters: Databricks is making governance more flexible and dynamic while enhancing data context, which is critical for Agentic AI.
How Securiti and Databricks are better together: In addition to structured data, Securiti addresses the specific challenges of unstructured data, namely that deep scans are necessary to classify data in those files at the elemental level. Securiti can map out and label all unstructured data across environments to inform policy as well as classify data as it’s ingested into DeltaTables. This provides the rich labelling needed for fine-grained controls that can be mapped to ABAC and governed via Unity Catalog.
Lakebase: Operational Database Layer
What’s New: Lakebase provides fast transactions using the systems companies already support by delivering data access for AI applications with Postgres compatibility and real-time sync on lakehouse tables.
Why it Matters: Lakebase bridges the gap between OLTP (think credit card swipes) and OLAP (think business reporting dashboards) creates friction between the operational and analytics hemispheres of the data estate. This is critical for Agentic AI that seeks to reduce the boundaries and latency between analysis and action but needs historical data, rich feature sets and fresh context in order to make good decisions and take optimal actions.
How Securiti and Databricks are better together: Operational applications and agents built on Lakebase can maintain permission boundaries from upstream sources. Securiti provides contextual insight about data that helps remediate lax or misaligned permissions. Meanwhile, many OLTP systems are stuffed with sensitive data, in unstructured data formats and loading to production AI in real-time. When Securiti syncs sanitized unstructured data to operational tables, the access context is preserved and can be applied to ensure applications and agents respect original entitlements.
AI/BI Genie: Conversational Analytics
What’s new: Genie allows natural language queries against data with explainable outputs based on curated business logic.Why it matters: Genie focuses on connecting business users and their data more easily. This allows developers to focus more on curating good, well governed data sets for business use and less on anticipating each specific business question that a user might ask.
How Securiti and Databricks are better together: Conversational queries through Genie will return results filtered by user permissions. By helping to better align permissions globally and preserving source system permissions, Securiti helps ensure that data retrieved by Genie for a given user is data that a user should have access to.
Databricks AI Security Framework (DASF) 2.0
What’s new: Databricks released an updated framework identifying 62 risks across 12 AI system components, and provides 64 controls to enable context-driven governance.
Why it matters: Databricks plays a central role in the data architectures of many of the world's biggest and most important organizations. The DASF 2.0 offers detailed and specific guidance to help secure AI systems.
How Securiti and Databricks are better together: Securiti offers a platform to implement the DASF 2.0 across data sources in the cloud and/or on prem. Automated data discovery and sanitization capabilities satisfy many of these controls. Data provenance, access control, and privacy protection are core features of the platform as well. Securiti provides a governance framework as well that allows for testing as well to ensure compliance with all policies and best practices.
What This Means Practically
For organizations using both platforms, the DAIS 2025 announcements create new opportunities:
- Expanded AI Use Cases: Teams can build more AI applications without worrying about permission leakage. The access control happens automatically based on source system entitlements.
- Faster Development: Developers don't need to build custom permission logic for each AI application. The infrastructure handles access control consistently.
- Simplified Governance: Instead of creating separate permission systems for AI applications, organizations extend existing access controls through the AI pipeline.
- Better Compliance: Audit trails connect AI outputs directly to source documents and permissions, making it easier to demonstrate compliance with data protection regulations.
Next Steps
If you're already using Databricks, the Securiti integration extends your existing investments by making unstructured data safely available for AI applications while maintaining permission boundaries. If you're using Securiti for data governance, the enhanced Databricks capabilities provide new ways to build AI applications on your sanitized, permission-aware datasets.
For teams evaluating both platforms, the integration provides a foundation for building AI applications that respect existing data access policies without requiring separate permission management systems.
Contact your Databricks or Securiti representative to discuss how these capabilities apply to your specific data governance and AI requirements.
WordsCharactersReading time
