PurposeEvidenceReviewApprovalOwnerConditionsMonitorAuditHuman controlPublic purpose
AI assurance for public institutions

Clear, accountable decisions about AI use

MIZAN brings together evidence, expert review and human approval for a specific AI use. It records who approved it, what the AI may do, the conditions attached, and when the decision needs review.

AI assurance means checking and documenting whether an AI use is fit for its purpose and its operating conditions.

In development. Exploring a focused design-partner pilot.

Illustrative concept using sample information. It does not represent a live deployment or final product interface.

Approval record snapshot

AI use
Resident service assistant (constructed)
Purpose
Explain published requirements and route complex cases to staff.
Accountable owner
Service delivery director
Decision
Authorised, with conditions
Review date
30 January 2027
Watch MIZAN

A short introduction to accountable AI approval

See how MIZAN frames evidence, authority and human responsibility around a specific AI use.

شاهد ميزان

مقدمة قصيرة عن الموافقة المسؤولة على استخدام الذكاء الاصطناعي

تعرّف على كيفية ربط ميزان بين الأدلة والصلاحية والمسؤولية البشرية حول استخدام محدد للذكاء الاصطناعي.

Approval record

What an accountable AI approval record contains

The approval record is the maintained artifact. It separates what the AI may do, what remains a human decision, which evidence was reviewed, and when the approval needs review.

Illustrative concept using sample information. It does not represent a live deployment or final product interface.

Resident service assistant approval

AI use
Resident-facing assistant for a public service.
Purpose
Explain published requirements, help residents prepare information, and route complex cases.
Accountable owner
Service delivery director.
Allowed
Use approved guidance, draft explanations, route cases to staff.
Not allowed
Make eligibility decisions, change records, or send final notices.
Evidence
Service policy, privacy and security review, accessibility testing, answer-quality review.
Decision
Authorised, with conditions
Conditions
Staff review for ambiguous cases; complaint signals trigger review.
Review date
30 January 2027.
Benefits

Why institutions use an approval record

Make responsibility clear.

Name the person responsible for the approval and the people who review the evidence.

Keep the basis for approval together.

Link the decision to the assessments, policies and conditions that support it.

Know when to review.

Flag changes that may affect the approval and route them for human review.

The problem

AI approval depends on more than a technical test

An AI system may perform well while important questions remain unanswered: who can approve its use, can the organisation supervise it, what must stay a human decision, and what happens if the system or its purpose changes.

Evidence spread across teams.

Policies, tests and reviews often sit in different documents and systems.

Unclear basis for approval.

A decision may be recorded without a clear link to its evidence, limits and owner.

Changes after launch.

A new model, policy, permission or incident may require the original decision to be reviewed.

People and public purpose

Keep people in control of consequential decisions

Public service AI should stay understandable, accessible and open to challenge. The approval record should define when staff must intervene, what decisions remain with people, and how concerns reach the responsible team.

Connected Abu Dhabi skyline representing public-purpose digital infrastructure.
Credibility and development

Applied research connected to practical delivery

MIZAN is being developed through Arrownex's AI for Public Purpose Lab in the UAE. The work combines institutional governance, privacy and public policy expertise with engineering and security delivery.

The methodology draws on public sector guidance and research on accountability, institutional capability and AI assurance. Pilot work will test whether the approach produces clearer decisions, reuses existing evidence and identifies when review is needed.

Arrownex

Delivery.

AI for Public Purpose Lab

Method.

UCL IIPP

Doctoral research, individual affiliation.

FAQ

Quick answers before you get in touch

Build the first case around your institution

Start with one AI use, one accountable owner and an agreed operating environment. Together, we can map the evidence, define the conditions for approval and agree how the pilot's value will be measured.

What we would agree: scope and responsibilities; evidence and review requirements; pilot deliverables and acceptance criteria.