# AI Workflow Risk Checklist

Use this before putting an AI workflow in front of real users.

## Data

- Data source approved
- Sensitive fields identified
- Retention assumptions documented
- Test data separated from production data

## Model Behavior

- Model/provider documented
- Prompt or tool behavior reviewed
- Failure modes listed
- Low-confidence fallback defined

## Human Review

- Review step visible
- Source evidence shown
- User can edit or reject output
- Approval history logged

## Security

- Roles and permissions defined
- API keys protected
- Logs do not expose unnecessary sensitive data
- Production access requires approval

## Operations

- Monitoring owner defined
- Escalation path known
- Handover notes prepared
- Next improvement backlog captured