Databricks databricks-aibi-dashboards — agentic threat model

Not certain from the listing — The specific foundation model (e.g., DBRX or external LLM) is not disclosed. Threats include prompt injection that could manipulate the model into generating malicious SQL queries or exposing underlying database schemas.

Directly binds to Lakehouse data to query and configure datasets. This creates a high risk of data exfiltration, unauthorized data access, or exposure of sensitive metadata if the agent is manipulated into querying unauthorized tables.

The agent framework translates user intent into dashboard schemas and dataset configurations. Vulnerabilities include insecure tool integration where the SQL generation tool might execute arbitrary commands or bypass query limits.

Not certain from the listing — While presumably hosted within the secure Databricks platform, the specific sandboxing of the agent's execution environment and network isolation boundaries are not detailed.

Not certain from the listing — Platform-level logging is expected, but specific guardrails to detect and block malicious query generation or anomalous data access patterns by the agent are not specified.

Crucial layer as the agent must respect Databricks Unity Catalog permissions and user IAM roles. A key threat is privilege escalation if the agent executes queries with higher privileges than the end-user.

Not certain from the listing — The agent is described as a standalone 'Agent Skill' within Databricks, with no explicit details on multi-agent collaboration or external ecosystem integrations.