Borrow a method from an examined AI case.
Compare bounded public-sector cases and return to the actual purpose, information, review work and owner. A sandbox finding is not transferable blanket permission.
- Lidingö and IMY: AI proposes redactions, a person owns disclosure: Follow the boundary between an AI masking suggestion and the official who owns disclosure.
- Talt: map the care workflow before building the AI service: Map the current process and separate a bounded test from operational use.
- Göteborg’s Copilot Chat assessment shows a real governance debate: Read a real municipal discussion of controls, legal caution and operational value without confusing Chat and Premium.
- AI in municipal HR: responsibility starts in the work: Bring the method back to Swedish public-sector guidance and a fresh local assessment.
A decision to discuss: Which conditions can we reuse, and which change when this becomes an HR workflow?