EU AI Act for public administration: duties as a deployer
Updated: 2026-09
Public bodies deploying high-risk AI face additional duties, notably the fundamental rights impact assessment. The first step is a complete register of the AI systems in use — which in most administrations does not yet exist.
AI has long been in use in public administration, often without being recorded as such: assistive features in departmental applications, text recognition, pre-sorting of requests, chatbots on the portal. These systems usually arrive as a feature of a procured product rather than as an AI project.
The EU AI Act reaches public bodies as deployers, and in certain constellations with additional duties beyond those of private deployers — notably the fundamental rights impact assessment for high-risk systems and notification to the supervisory authority.
The practical problem is therefore not the assessment but the inventory. A body that does not know which AI runs in the house can neither classify nor assess. And that knowledge sits spread across departments and procurement files.
How it works
- 01
Collect the inventory across departments
Ask systematically which applications contain AI features — including where they arrived as an add-on feature of a procured product.
- 02
Establish provenance per system
Record whether the system was built, procured or delivered as a feature. The provider or deployer role follows from that.
- 03
Classify under Article 6 and Annex III
Establish which systems fall into the high-risk category. Public-administration use cases appear repeatedly in Annex III.
- 04
Plan the fundamental rights impact assessment
For high-risk systems, schedule the FRIA and factor in notification to the supervisory authority.
- 05
Make human oversight real
Document which person can review and override a decision — and whether they understand the system well enough to do so.
- 06
Keep the register maintainable
Tie classification to the procurement process, so new systems do not again arrive unnoticed.
Why it fits here
Team workspaces mirror the distributed reality of an administration: departments enter their own systems while the overall view stays central.
Audit-ready PDF reports are the format a supervisory or audit review actually asks for.
Classification with legal citations makes it traceable why a system was assessed as it was — in an administration the reasoning counts for more than the result.
FAQ
Does the AI Act apply to Swiss cantons and municipalities?
The AI Act is EU law. For Swiss bodies, relevance arises mainly from EU links, procurement requirements and partner expectations. Actual applicability belongs clarified case by case.
What is a fundamental rights impact assessment?
A structured assessment of a high-risk AI system’s effects on fundamental rights. Public bodies deploying such systems are obliged to produce one and to notify the supervisory authority.
We only have AI as a feature in procured software.
That is the most common case — and it does not change the deployer role. Classification follows the system and its purpose, not how it entered the house.
Where do we start?
With the inventory. Without a complete register of the AI systems in use, all further work is piecemeal.
Does this replace legal advice?
No. The tool delivers inventory, classification and evidence. Legal assessment stays with the responsible body and its advisers.
AI Risk Check
Team workspaces mirror the distributed reality of an administration: departments enter their own systems while the overall view stays central.
Audit-ready PDF reports are the format a supervisory or audit review actually asks for.
Classification with legal citations makes it traceable why a system was assessed as it was — in an administration the reasoning counts for more than the result.