# Risk traffic light

## Green — proceed and verify

Public, synthetic or approved low-sensitivity inputs; reversible output; clear human review.

## Amber — pause and add controls

Internal context, code changes, customer-impacting recommendations, ambiguous permissions or material business consequence.

Required controls:

- Reduce data/action scope
- Name accountable reviewer
- Add fixed evaluation or test
- Preserve sources and audit trail
- Require approval before write/release

## Red — stop and escalate

Secrets, credentials, unapproved personal data, autonomous production action, unsupported legal/compliance claim or a task outside approved access.

## Team decision

- Scenario:
- Classification:
- Data involved:
- Action involved:
- Potential consequence:
- Required human:
- Evidence needed:
- Escalation owner: