Record a humanitarian AI task

Use an editable record to connect the task, evidence, separate assessments and human decision.

Draft guidance. Classification does not assign a risk rating or authorise use.

Download the record

Download the editable Word template

Download the completed translation example

Use one record for each defined task. Save the Word file in your organisation’s approved system. This website does not collect or save your answers.

Complete it in three passes

  1. Describe the task. Record its purpose, setting, inputs and result. Explain the capability and function codes you choose, then specify what the system is permitted to do.
  2. Check people and evidence. Describe who could be affected, what an error could change and what evidence supports the proposed use. Record who contributed and who was missing.
  3. Record assessments and the human decision. Link the DCI and SAFE AI assessments separately, with their status and version. Record the authorised person’s decision only when one exists. Identify monitoring, fallback and stopping arrangements.

Keep uncertainty visible

Use “Unknown” when a fact is missing, “Not assessed” when an assessment has not been completed, and “Not applicable” only with a reason. A blank field is ambiguous. Do not turn missing evidence into a low-risk rating or assume that a named reviewer has approved use.

Keep sensitive evidence in its designated system and reference it here. Do not paste beneficiary details or confidential source material into the record.

Review and maintain the record

Ask a second reviewer to check the classification and its reasoning. Record disagreement and unresolved questions. Retain the earlier version when the task, system, setting or permissions change, then link the revised record to any reassessment.

The completed translation example demonstrates the fields. It deliberately marks operational facts and assessment results as unknown or not assessed. It does not demonstrate an approved deployment.

Sources

A. A Taxonomy for AI in Social Protection: Defining Capabilities and Application Areas

B. AI Hub Risk Assessment Framework for Social Protection: A Practical Guide to Responsible AI Deployment

F. SAFE AI Governance Framework