AIdå

In the public sector, the question "can we use this AI service?" is asked far more often than it is answered. The answer depends on information classification, personal data, agreements and the safeguards the service actually provides — and it usually lands on an already overloaded data protection or information security function.

I wanted to explore whether that first assessment could be turned into a tool: simple enough to be used by a systems manager or project lead, careful enough to never become a data protection problem itself.

AIdå is an iOS app that guides the user to a clear verdict on a planned AI use:

The assessment is based on what the user indicates about the situation — the type of data involved, confidentiality and accuracy requirements, the chosen service and how it is run — never on the content itself. The app accepts no free text and collects no personal data. It works offline, with a bundled catalogue of AI services and their documented safeguards.

It would have been easy to make AIdå another chatbot. Instead, the decision engine is deterministic: rules, gates and priorities in plain language. If personal data is involved and no data processing agreement exists, it does not matter how good the service is otherwise — the verdict is "requires an agreement". Sensitive data requires the service to demonstrate data residency within the EEA, encryption and a commitment not to train on customer data.

Every verdict comes with its reasoning: which rules were triggered, and what it would take to reach an OK. That makes the advice reviewable — the same property I argue for in my article on AI agents and deterministic systems.

AIdå is a Swedish pun that works in both directions: "AI då?" — "what about AI?", the question you ask before starting — and "aj då", roughly "uh-oh", the reaction when the answer is a discouragement. The prototype's tagline: "Kanske inte den." — "Maybe not that one."

AIdå is a working prototype, built in SwiftUI with a decision engine, a service catalogue and export of the assessment — but it is not a released product. It was built to test whether the assessment logic holds, not to be launched.

The idea lives on in service form: the same questions, gates and requirements are now the backbone of my AI Impact Assessment (AIIA). What the app does in milliseconds, I do as an assessment — grounded in the organization's actual context, agreements and risks.