An older man reads a letter at a bright kitchen table, trying to make sense of it.

← Back to overview

The Mandated Assistant

Accountability.

Illustrative

Before

Before

Advisers sit with a presenter as a plan is walked through, before anyone asks about risk.
Photo: Mikhail Nilov / Pexels

Trigger

Advice without a risk policy

An external consultancy advises the ministry to introduce an AI verification assistant, framed as efficient and low-risk. No risk maps or safety frameworks are created.

Hands crowd over laptops and marked-up pages as a deadline closes in.
Photo: Scott Graham / Unsplash

Process

System drift

The project is led for speed. AI risks are deprioritised, roles are unclear, and there is no escalation route.

An older man reads a letter at a bright kitchen table, trying to make sense of it.
Photo: T Leish / Pexels

Result

Confusion and distrust

Older users feel confused and excluded and become more vulnerable to scams. The public sees the system as driven by hype.

Proposed

What the role paths change

After (speculative)

After

A mixed group sits around one table, looking at a plan together.
Photo: fauxels / Pexels

Trigger

Governance first

The consultancy and the ministry agree a governance-first approach, aligned with Verifiland's AI law and data protection.

A small team looks through papers and a tablet, checking the work item by item.
Photo: fauxels / Pexels

Process

Checkpoints at every stage

VTS leads development with risk mapping, human oversight and compliance checkpoints at every stage.

An older woman smiles as she uses a tablet, comfortable with the step in front of her.
Photo: Kampus Production / Pexels

Result

Taken seriously

Older citizens feel confident using the assistant. It is clear and transparent, and they feel taken seriously.

The before row is drawn out on purpose. The after row is deliberately hopeful.

From the interviews

  • There is no formal framework for non-technical or long-term risks.

    Innovation developer. Interview insight, paraphrased

  • There is pressure to show impressive results, but no clear way to assess AI risks.

    Innovation manager. Interview insight, paraphrased

Where it goes wrong, and where the path steps in

Where it goes wrong, and where the path steps in Before After Strategy ✕ ✓ Amira · Product Owner Commission ✕ ✓ Amira · Product Owner Design ✕ ✓ Lena · UX/UI Designer Build ✕ ✓ Jannik · Developer Operate ✕ ✓ Sabine · Caseworker Citizen ✕ ✓ Sabine · Caseworker

Guardrails used

  • Risk logging and traceability
  • Ethics review
  • Human-in-the-loop controls

Checklist aligned with the risk functions: Map · Measure · Manage · Govern

What each employee does

Sabine Krüger Caseworker

Signs a citizen is confused or at risk of scams.

Flags confusing AI answers from real calls.

Escalates the cases that are at risk.

Real examples to Lena, the UX/UI designer, and Amira, the product owner.

Lena Richter UX/UI Designer

Plain language for uncertain states.

Rewrites it as “still being checked”, with a next step.

Tests the wording with an affected user.

Revised message.

Jannik Voigt Developer

Mapping statuses to allowed messages.

Blocks “ineligible” unless the status is final.

Adds tests for pending, missing and failed statuses.

Test results.

Amira Haddad Product Owner

What a governance-first AI project needs before it starts.

Asks the advisers for a risk plan before agreeing the scope.

Names the reviewer and escalation route in the team agreement.

A signed-off agreement to the team and the ministry.