What the AI Act means for Swedish municipalities
The EU AI Act sets new, strict requirements for how artificial intelligence may be used in the public sector. Here is what municipal decision-makers, IT managers and heads of service need to know before procurement and deployment.
The AI Act’s risk-based framework
The EU AI Act divides AI systems into four risk levels: unacceptable risk, high risk, limited risk and minimal risk. Public administration, and in particular services that decide on individuals’ rights and assistance (such as social services and education), is a focus of the legislator’s scrutiny.
A common misconception is that all AI in a municipality is automatically considered prohibited or subject to insurmountable regulation. The legislation does not aim to stop digitalisation, but to ensure transparency, traceability and human control.
Human in the loop
The key requirement of the AI Act for the public sector is the principle of “human in the loop”. An AI system may never independently make official decisions or carry out automated legal measures that adversely affect citizens.
Interpreto is designed from the ground up as administrative support and a system for preparing case material. This means that the AI transcribes, summarises and suggests text based on conversations and policy documents, but it is always the authorised caseworker who reviews, edits and signs the decision. This clear division of responsibility makes the system fully compatible with the regulation’s requirements.
Data sovereignty and the ban on US clouds
In addition to the AI Act, Swedish municipalities must comply with the Public Access to Information and Secrecy Act (OSL), GDPR and the Schrems II ruling. When sensitive personal data about children, victims of crime or recipients of assistance is processed, the data must not be exposed to foreign legislation such as the US Cloud Act or FISA 702.
Meeting these requirements means that all data processing, transcription and inference must take place exclusively within the borders of Sweden and the EU, on sovereign European infrastructure, without US dependencies.
Checklist for municipal AI procurement
1. Make sure the supplier does not train its foundation models on your municipality’s data.
2. Require full data sovereignty, with operation in certified Swedish/European data centres.
3. Verify that the solution works as support (human in the loop) and not as an automated decision-maker.
4. Check that logging and traceability are in place for every piece of material generated.
Summary
The AI Act sets clear requirements for transparency and human control. By choosing solutions operated in Sweden, with a human at the helm, municipalities can modernise their services with confidence and in full compliance with EU law.
Want to try it in your organisation?
Book a walkthrough and see how your templates can be automated.