Security & Compliance understanding on LLM Usage in Databricks Genie and Agentbricks

[abhijit007]

Hi everyone,

With the increasing focus on security and compliance for AI Agents and LLMs, I wanted to get some clarity on a couple of points related to Databricks Genie and Agentbricks.

Could someone help provide detailed information on the following, along with references to any official Databricks documentation or best practices?

1. Which LLMs are used by Genie and Agentbricks (including the Supervisor Agent)?
   I want to understand what underlying models are powering these components.
2. How does Databricks ensure data security when Genie and Agentbricks interact with user data?
   Since these agents leverage LLMs behind the scenes, I’d like to know what mechanisms protect the data being passed to or processed by the models.

Any insights or pointers to relevant Databricks resources would be greatly appreciated.

Thank you!

[Ashwin_DSA · Accepted solution]

Hi @abhijit007,

Please take a look at these pages. They answer your queries in detail for Genie.

• https://docs.databricks.com/genie - Covers architecture and how it works. Also covers security.  
• https://docs.databricks.com/databricks-ai/databricks-ai-trust - Covers data handling, training, encryption and residency
• https://docs.databricks.com/databricks-ai/partner-powered - Partner-powered AI features (which models power which features)
• https://docs.databricks.com/resources/designated-services - Databricks Designated Services (model provider + Geo availability)

And for agent bricks, refer to the below.

• https://docs.databricks.com/generative-ai/agent-bricks - Architecture
• https://docs.databricks.com/generative-ai/agent-bricks/multi-agent-supervisor - Storage, requirements etc.
• https://docs.databricks.com/generative-ai/mcp/external-mcp - External MCP security & authentication
• https://docs.databricks.com/resources/designated-services - Services & residency

Does this help? If you have any specific questions, please do ask.

If this answer resolves your question, could you mark it as “Accept as Solution”? That helps other users quickly find the correct fix.

[abhijit007]

Hi @Ashwin_DSA ,

Thanks for the details it's really helpful regarding security and compliance... But could you please help me on the first point?

1. Which LLMs are used by Genie and Agentbricks (including the Supervisor Agent)?
   I want to understand what underlying models are powering these components.

[Ashwin_DSA · Accepted solution]

Hi @abhijit007,

Both Genie and Agent Bricks are built as managed, model‑flexible services rather than being tied to a single fixed LLM.

Genie is implemented as a compound AI system that uses LLMs plus Unity Catalog metadata, example SQL, and space instructions to translate natural language into SQL and answers. When partner-powered AI features are enabled, Genie uses models hosted by Azure OpenAI / Azure AI Services as the underlying LLM provider. Databricks can upgrade or change the specific base model over time as part of the managed service.

Agent Bricks (including Supervisor Agent) is model‑agnostic. It uses Mosaic AI Model Serving and AI Gateway to work with foundation models available in Unity Catalog (the system.ai schema) or other configured model endpoints. Supervisor Agent also requires access to foundation models in system.ai and the databricks-gte-large-en embedding model. The supervisor logic itself runs on Databricks‑hosted foundation models, and the exact model choice can evolve over time as Databricks optimizes quality and performance.

There's more detailed information in the links I shared earlier.

Does that answer your question?

If this answer resolves your question, could you mark it as “Accept as Solution”? That helps other users quickly find the correct fix.

[abhijit007]

Hi @Ashwin_DSA ,

It helps.. Thanks for the details.