AI Security at Scale: Lessons From Securing Databricks and Our Customers

Securing AI at scale is challenging. This session provides a dual view: how Databricks secures its own AI infrastructure, handling millions of daily queries and patterns learned from customers using Databricks AI features.

Our discussion begins with how Databricks secures its AI offerings — Agent Bricks, AI/BI Genie, and Model Serving. Key to this security framework are architectural choices, including the use of Unity Catalog for identity propagation and fine-grained access control, and the implementation of safeguards to defend against the “Agentic AI Lethal Trifecta.”

We then cover five major AI security challenges faced by Databricks customers and show how the Databricks platform mitigates these risks. These include AI Asset Surface Sprawl, Sensitive Data Leakage, Over-Privileged AI Systems, Prompt Injection and Observability Gaps.

Through this presentation attendees will gain actionable frameworks, architectures, and patterns for securing production AI at scale.