Sweden's data authority puts AI in the public sector under the microscope

IMY turns its attention to AI in the public sector

Sweden's data protection authority IMY has made AI in the public sector one of its priorities for the coming years. In its supervision plan for 2025 to 2027, published on 14 October 2026, the authority says it will inspect and guide public bodies that use AI to process sensitive personal data, and where citizens have limited freedom of choice ^[1].

The plan doesn't name specific authorities or systems. Instead, it signals where IMY will look: at AI that affects people in situations where they can't simply walk away.

Think of it like a canteen. If your employer has one, you can bring your own lunch. If you're in custody, you eat what you're given. When an authority uses AI to decide something about you, the power imbalance is closer to the second case. That's why the scrutiny matters.

What IMY will inspect

IMY's plan mentions three areas ^[1]:

• AI systems that process sensitive personal data, such as health information, biometric data or data about criminal convictions.
• Situations where citizens have little or no choice, for example when an authority is the only body that can grant a benefit or a permit.
• Follow-up on earlier orders and guidance, particularly around automated decision-making under Article 22 of the GDPR.

The supervision plan is a signal, not a list of cases. It tells authorities what IMY will be watching, so they can get their house in order before an inspection arrives.

The AI Act's public sector rules

From 2 December 2027, public bodies that use high-risk AI systems must follow Article 26 of the AI Act ^[2]. One part of that article is often overlooked: before a high-risk system is put into use in the workplace, the employer must inform trade union representatives and employees (Article 26(7)).

That rule applies whether the AI is used on the authority's own staff or on the public. The point is transparency: people who will be affected get to know that the system is coming.

Public authorities are also among the users who must carry out a fundamental rights impact assessment under Article 27 before they start using a high-risk AI system ^[2]. The assessment covers risks to rights such as privacy, non-discrimination and fair trial.

Logs and decisions under pressure

High-risk AI systems must allow for the automatic recording of events over their lifetime, under Article 12 ^[3]. That applies from 2 December 2027 for systems in Annex III, which includes several areas of public administration ^[4].

The law says what to log. It doesn't say how to prove the log hasn't been edited.

That gap matters in the public sector. A decision about benefits, residence or child protection can be questioned months or years later. If the log that explains the decision sits on a server the authority controls, and the authority could have changed it, the record becomes easier to challenge.

Sealing decisions as they are made

This is where a digital witness fits. The thing to seal is each decision record as it is written to the log: the case reference, the model version, the input summary, the output and the reviewing officer. Sealing takes a fingerprint of the record, not the record itself, so the decision and any personal data never leave the authority's systems.

When a decision is questioned later, the authority can show the sealed record to an auditor, a supervisor or a court. Anyone it shares the record with can check that the fingerprint still matches. If it does, the record is exactly what was written on the day. Change a single character and it won't match.

Sealing happens as the record is written, through WitniumChain, using the REST API, the SDK or the MCP server. The first 2,000 witnesses a month are free. Technical documentation is at witniumchain.com/docs.

What sealing doesn't do: it doesn't show that the decision was lawful, fair or correct. It shows that the record of what the system did, and when, hasn't changed since it was sealed. That's a narrower claim, and a checkable one.

How it would work

1. The authority's case management system writes a decision record to the log.
2. The system calculates a fingerprint and sends it to WitniumChain, getting a witness ID back.
3. Months later, when the decision is reviewed, the authority recalculates the fingerprint and compares it with the sealed one. If they match, nothing has changed.

What this means for you

If you work in a public authority that uses or is planning to use AI, IMY's supervision plan is worth reading in full ^[1]. Here are three questions to ask now:

• Does the AI process sensitive personal data, as defined in Article 9 of the GDPR?
• Have we carried out a fundamental rights impact assessment, if we're required to under Article 27 of the AI Act?
• If a decision from our AI system were questioned a year from now, could we show what the system did, with a record nobody could have quietly edited?

The third question is the evidence question. It's easier to answer before the first inspection than during it.

What to watch

Two dates stand out. From 2 August 2027, general-purpose AI models with systemic risk must comply with the AI Act's transparency and risk management rules ^[5]. From 2 December 2027, the high-risk requirements in Annex III of the AI Act apply, including the logging duty in Article 12 ^[4].

IMY's supervision plan runs to 2027, so expect guidance and inspections in the lead-up to those dates.