AI Workflow Risk Check

A practical screen for authority, data, failure and oversight risks before an AI workflow reaches live operations.

The workflow has one named accountable owner.

AI permissions are limited to the minimum required.

Sensitive data use and retention have been reviewed.

Consequential actions require proportionate human approval.

Significant actions are logged and reviewable.

Failures and provider outages have defined handling.

Unexpected cases reach a named human owner.

The workflow can be stopped without relying on the AI itself.

The proposed control model covers the essential areas. Verify that the documented controls operate in real tests.

The workflow has material governance gaps that should be resolved before authority or data access expands.

The workflow should not enter live operations in its current form. Reduce authority and design ownership, observation and shutdown first.

• Controlled AI Operations
• All diagnostic tools
• Contact Four Gears