Power BI AI Lockdown Modes

Let teams use AI for Power BI
without giving AI unrestricted model access

SemanticOps lockdown modes define what assistants and users can inspect, query, modify, or block across Power BI semantic models by environment, team, and risk level.

The problem

AI access is not binary

AI tools typically operate with the permissions of the user running them. In enterprise environments, that is too broad.

A developer may have edit rights. An AI assistant may have tool access. A production semantic model may have no contextual safeguards. Different environments, teams, and models need different permission levels — and those levels should be centrally enforceable, not left to each user to configure.

Available modes

Seven lockdown modes for different risk levels

Match the mode to the environment — local development, shared staging, or a locked production model.

Full access

Local development and experimentation

All model operations available. Intended for isolated developer environments where speed matters more than restriction.

Guarded access

Shared development with policy checks

Full operations with confirmation prompts and policy enforcement. Prevents unsafe changes while preserving developer velocity.

Read-only

AI can inspect and explain, but not modify

AI assistants can browse the model, read metadata, and run non-data queries. All write operations are blocked.

Browse-only

AI can navigate metadata with minimal exposure

The most restricted mode for AI use. Limited to metadata navigation with no query or modification access.

Test-only

AI can run validation but cannot modify

Allows the AI to execute test suites and read results without the ability to change model metadata.

Production lockdown

Writes, deletes, and sensitive operations blocked

Designed for production environments. All destructive and data-exposing operations are blocked regardless of user permissions.

Enterprise enforced

Admin policy cannot be bypassed locally

Policy bundle is loaded from a central path. Individual users cannot override the mode or relax the rules.

Controls

What can be restricted

Lockdown modes control access at the operation level — not just read/write. Teams can restrict data-returning queries without blocking metadata reads, or allow test execution while preventing model changes.

Model browsing

DAX execution

Data-returning queries

Metadata reads

Measure creation

Object rename

Object delete

Relationship changes

RLS / OLS changes

Bulk operations

Service / XMLA connectivity

Audit log exposure

Data masking controls

Policy bundle path

Policy fail mode

Enterprise deployment

Central policy enforcement that users cannot bypass

In enterprise mode, the policy bundle is loaded from an admin-controlled path. Users cannot override the lockdown mode or modify which policies apply. This ensures consistent enforcement across all developers, environments, and AI clients in the organization.

Admin sets the policy bundle path

The admin deploys the bundle; all local installs pick it up automatically.

Mode cannot be overridden locally

Users cannot relax the lockdown level for their own session.

Consistent enforcement across all clients

Claude Code, VS Code, Cursor, and any other MCP client operates under the same rules.

Design an AI access model for Power BI.

Match permissions to environments. Lock down production. Let developers move fast in isolation.