# Risk and review matrix

Use this matrix to decide how much evidence, review, and testing an AI-assisted workflow requires. Its score bands are an author-proposed starting heuristic, not a validated risk instrument. Calibrate the factors and routing rules to real assets before adoption. Score consequence and exposure before generation, and do not lower the tier because a deadline is close.

## Project classification

- **Project or asset:**
- **Intended audience and use:**
- **Target performance:**
- **Accountable owner:**
- **Classifier and date:**

## Risk factors

| Factor | Low — 1 | Moderate — 2 | High — 3 | Score | Reason / evidence |
|---|---|---|---|---:|---|
| Consequence of error | Easily noticed and corrected; little effect on decisions | Could cause rework, service failure, or a consequential misunderstanding | Could affect safety, rights, employment, qualification, finances, compliance, or essential access | | |
| Source uncertainty | Current, approved, consistent sources | Gaps or interpretation questions with assigned owners | Missing, conflicting, obsolete, or unowned controlling sources | | |
| Data sensitivity | Public or approved nonsensitive information | Internal or proprietary material with explicit controls | Personal, confidential, regulated, restricted, or unapproved information | | |
| Audience scale | Small, known group with direct support | Repeated use across a function or organization | Public, high-volume, difficult-to-recall distribution | | |
| Assessment impact | Practice or informal self-check | Completion, proficiency, or manager decision | Certification, selection, disciplinary, eligibility, or other high-impact use | | |
| Accessibility exposure | Simple accessible format with established test path | Multiple media or interactions requiring manual testing | Essential task depends on complex interaction or an untested alternative | | |
| Rights and reuse | Original or clearly licensed sources and outputs | Mixed ownership or external assets needing confirmation | Unclear rights, protected material, biometric likeness, or restricted reuse | | |
| Maintenance volatility | Stable content with a named owner | Periodic updates or vendor-dependent behavior | Fast-changing rules, product behavior, or no maintenance owner | | |

## Provisional tier

Add the eight scores.

- **8–11 — Tier 1:** controlled production risk;
- **12–17 — Tier 2:** elevated review risk;
- **18–24 — Tier 3:** high-consequence or high-uncertainty use.

A score of 3 for data sensitivity, assessment impact, or consequence of error moves the workflow to at least Tier 2. An unresolved stop condition overrides the total.

## Required controls

| Control | Tier 1 | Tier 2 | Tier 3 | Assigned owner / evidence |
|---|---|---|---|---|
| Versioned source pack | Required | Required | Required with independent authority check | |
| Documented AI-use approval | Per organizational policy | Required | Required from designated governance owner | |
| Claim verification | Material claims | Full high-consequence content | Full traceability and independent sampling | |
| SME review | Named content owner | Required specialist review | Independent or multi-role review as assigned | |
| Learning-design review | Required | Required | Required with formal alignment evidence | |
| Privacy, legal, policy, or security review | If triggered | Required for relevant risks | Required before generation and release | |
| Accessibility testing | Format-appropriate checks | Automated and manual checks | Specialist and user testing appropriate to context | |
| Representative learner testing | Recommended | Required sample | Formal plan appropriate to consequence | |
| Release owner | Required | Required | Required with documented risk acceptance | |
| Maintenance and withdrawal plan | Named owner | Version, review date, and withdrawal path | Monitoring, escalation, recall, and audit record | |

## Stop conditions

- [ ] A controlling source is missing, obsolete, or unresolved.
- [ ] The intended data is not approved for the selected system or purpose.
- [ ] No qualified person owns a high-impact decision.
- [ ] The planned assessment will influence a consequential decision without the required specialist review.
- [ ] The essential learner task cannot be accessed or tested in the delivery environment.
- [ ] Rights or permissions are unresolved.
- [ ] No one can maintain, correct, or withdraw the released asset.

If any checked condition remains unresolved, stop the affected workflow.

## Classification decision

- **Tier:** 1 / 2 / 3 / Stop
- **Required reviewers:**
- **Required evidence:**
- **Restricted inputs or uses:**
- **Approval owner:**
- **Decision date:**
- **Reclassification trigger:**
