# AI Boundary Decision Worksheet

Use this before building or deploying an AI feature. The goal is to define what the AI is allowed to do, when humans must review it, and what happens when behavior is uncertain or wrong.

## AI Role

- Product or feature:
- Intended user:
- Intended setting:
- AI output:
- Is the output advisory, administrative, triage-supporting, or decision-driving?
- What human decision or workflow does it influence?

## Boundary Decisions

| Decision | Current answer | Evidence or rationale |
| --- | --- | --- |
| What is the AI allowed to do? |  |  |
| What is it not allowed to do? |  |  |
| Who reviews output before action? |  |  |
| What triggers mandatory review? |  |  |
| What happens when confidence is low? |  |  |
| What happens if the AI is unavailable? |  |  |
| What output should never be shown to users? |  |  |

## People and Populations

| Group or setting | Possible performance concern | How it will be checked |
| --- | --- | --- |
|  |  |  |

## Monitoring and Change

- Performance indicators to monitor:
- Subgroup checks:
- Complaint or feedback signals:
- Drift triggers:
- Rollback triggers:
- Owner for monitoring:
- Owner for approving model or prompt changes:

## Go / No-Go

- [ ] Intended use and non-use conditions are written.
- [ ] Human review boundary is clear.
- [ ] Known limitations are visible to the right people.
- [ ] Data provenance and evaluation data are documented.
- [ ] Rollback or disable path exists.
- [ ] The team can explain what happens when the AI is wrong.