Illustrative scenario
An assistant translates an approved notice for public distribution. Use H-COMMUNICATE and C-GEN. The consequences depend on the notice: an error about where, when or how to obtain help could change what someone does. A competent language reviewer checks its meaning and accessibility. Record the approved text, reviewer and permission to publish. Permission for the assistant to send the notice must be recorded separately.
Completed task record
Download the completed Word example. All entries below describe an invented proposal. No checks, assessments or authorisation have been completed.
Purpose and setting
Record and version. Illustrative record TRANSLATION-DEMO. Version 1, 25 September 2026. Not an operational record.
Task and purpose. Draft a translation of an already approved public service notice. Intended result: an accurate, accessible notice. Alternative: translation by a competent person without AI. No benefit has been measured.
Owner and setting. Proposed owner: communications lead. No person assigned. General service communication. Location, language pair and operational conditions are unknown and must be specified before assessment.
Classification
Humanitarian function. H-COMMUNICATE. The task communicates service information to people. No other function is assigned to this defined task.
Steps and capabilities. Input: approved source notice. AI step: draft a translation. Result: candidate translated text. C-GEN applies because the system transforms text into another language. Human checking and publication are separate steps.
System and permissions
System and evidence. System and version not selected. Performance, data handling and language quality have not been tested.
Authority boundary. Proposed boundary: read the approved notice and draft a translation. No contact lists, case records or sending tools. Stop after drafting. A person must check the translation and a separately authorised person must approve publication.
People and consequences
People affected. Communications staff use the draft. People seeking service information may rely on the notice. People using minority languages or accessible formats may be poorly served.
Consequences of error. Incorrect locations, dates, eligibility wording or instructions could delay access to help or cause an unnecessary journey. A fluent translation can still change meaning. Severity depends on the actual notice and setting.
Data and evidence
Sources and restrictions. No actual notice or evidence is attached. Use an approved source notice and check that it contains no personal or restricted information before supplying it. Storage, retention and provider terms remain unverified.
Checks and gaps. No checks performed. Before publication, a competent language reviewer should compare the draft with the approved source, including meaning, service details, accessibility and harmful ambiguity. Record corrections and review evidence in the organisation’s approved system.
Participation
Who contributed. No consultation has occurred. Seek input from intended readers on language and accessible formats through an appropriate process. Record who was missing and any disagreement without exposing respondent identities.
Separate assessments
DCI assessment. Not assessed. No DCI rating assigned. Applicability and the relevant assessment process must be determined for the actual use.
SAFE AI assessment. Not assessed. No SAFE AI tier assigned. Assess the notice, audience and consequences in context. The taxonomy codes alone cannot establish a tier.
Other required reviews. Not identified. Check the organisation’s information, data protection, accessibility and approval requirements for the actual setting.
Human decision
Decision and authority. No decision taken. No reviewer or decision-maker assigned. This example provides no authorisation to process data or publish a notice.
Conditions and unresolved matters. Unresolved: actual source notice, languages, system, evidence of translation quality, affected readers, assessment outcomes and authorised roles. Record these before any operational decision.
Operation and review
Monitoring and fallback. Proposed approach: a responsible communications role receives error reports and maintains the corrected notice. Use a competent human translator if the AI result cannot be verified. No arrangement is established.
Stopping and reassessment. Suspension authority not assigned. Reassess if the notice, audience, language, system or permissions change, or if an error could affect access to help. Retain earlier records and explain changes.
Review questions
Ask another reviewer to classify the task independently. Compare the codes and explain any disagreement. Identify the data, permissions and consequences that would need checking in a real setting.
Status
This is an invented AidGPT example. No deployment, measured benefit, DCI rating, SAFE AI tier or approval is claimed.
Sources
A. A Taxonomy for AI in Social Protection: Defining Capabilities and Application Areas