New AI rules for public authorities and other organisations. We help you move forward.

By way of illustration: this blog post was written with the help of AI, and subsequently verified and adjusted by the Nazka Mapps team. The fact that we are telling you this is no coincidence — it is exactly what article 50 of the AI Act will require from public authorities as of 2 August. Read on and we will explain why.

What changes on 2 August?

From 2 August 2026, article 50 of the AI Act applies: the transparency obligation. In short, citizens must henceforth know when they are dealing with AI:

• Chatbots and digital assistants: if a citizen interacts with an AI chatbot via your website, this must be clearly indicated.
• AI-generated content: texts, images, audio or video that are (partly) produced by AI and inform the public about matters of general interest must be clearly labelled as AI-generated.
• Deepfakes and misleading content: where there is a risk of deception, a label is mandatory, regardless of the purpose.

For Flemish public authorities this is not an entirely new world: the principle of open and correct communication with citizens already existed. Article 50 now makes it explicit for situations in which AI is used. In case of doubt, the AI Expertise Centre of the Flemish government (ai@vlaanderen.be) answers questions on this topic.

The Belgian government is ahead — and that requires attention

Strikingly, while we often think that companies form the technological vanguard, we see the opposite in Belgium. Research by Aceerta shows that the public sector adopts AI significantly faster than the private sector, with usage at 75.8% among public employees versus 62% in the private sector. Public authorities are actively integrating AI into their core services: from digital front offices to data analysis for policy preparation.

This means that the transparency obligation of 2 August will be a very concrete test for public organisations. Those who deploy AI for citizen communication, reports or map visualisations must be able to demonstrate where and how that happens.

Where is the risk for geodata and maps?

For map platforms and geodata applications, the impact depends on how they are used:

• Interactive maps and dashboards without an AI component fall outside the strictest obligations — you design and manage them as before.
• AI-based analyses or generated map texts (for example, automatically generated descriptions for climate risk maps or summaries of sensor data) must be identifiable and labelable as AI output.
• High-risk applications — think of AI that supports spatial decisions about individuals, enforcement or critical infrastructure — will only face the full weight of conformity assessments, logging and human oversight later (December 2027).

High risk: when maps start steering decisions about people

Most geodata projects are low-risk. But as soon as AI-based spatial analyses are used to support decisions about individuals, enforcement or vital infrastructure, an application moves towards the high-risk category. Think of: AI that determines which neighbourhoods receive extra inspections, systems that allocate subsidies or services based on location and profile, or predictive analyses for police deployment. For these applications, from the end of 2027 stringent requirements apply regarding data quality, logging, human oversight and a fundamental rights impact assessment.

The classic pitfall example: bias in spatial data.

Suppose a municipality uses AI to predict which homes have the highest risk of energy poverty, based on historical renovation and subsidy data. If certain neighbourhoods were systematically measured less in the past, applied for fewer subsidies (for example due to a digital divide or language barriers), or received fewer inspections, then the model will "learn" that blind spot as well. The result: the map appears to show objective risk zones, but in practice diverts resources away from exactly the most vulnerable districts — and thereby reinforces an existing inequality.

This is precisely why the AI Act requires that, for high-risk systems, training and input data must be relevant, representative and as complete as possible, and that known bias is investigated and documented. For geodata there is an extra dimension: spatial representativeness. Are all sub-municipalities equally well covered? Is sensor data fairly distributed across poorer and wealthier neighbourhoods? Are the coordinates of incident reports reliable, or do they mainly come from districts where people report more often? These are questions no algorithm answers by itself — they require domain and data expertise.

What can Nazka Mapps do for you?

We are not lawyers, but we are your technical partner for responsible geodata and AI. Concretely:

• Transparent map interfaces: we build labelling and provenance information directly into your interactive maps, so citizens can see where AI has been used and where not.
• Data quality as a foundation — also against bias: AI output is only reliable if the underlying geodata is correct, up to date and spatially representative. We check datasets for coverage and distortion and document this, so that during an audit or FRIA you can demonstrate where your data comes from — and where the limitations lie.
• Building in human control: our platforms are designed with a "human-in-the-loop" mindset — civil servants retain final responsibility for spatial decisions.
• Preparing for high risk (2027): for projects that are moving towards high risk, we help you with data bias audits, logging architecture and preparing the mandatory fundamental rights impact assessment (FRIA). Those who get their data foundations in order today will not have to scramble in panic in 2027.

The AI Act does not have to be a brake. With the right technical foundation, it becomes an opportunity to build trust: among citizens, within your organisation and with regulators.

Do you want to know what 2 August specifically means for your map applications or geodata projects? Contact us — we are happy to think along.

About this blog post: this text was drafted with the help of AI and subsequently verified in substance by the Nazka Mapps team. We believe you should know that — and from 2 August, the AI Act agrees.

Sources:

1.      https://www.vlaanderen.be/digitaal-vlaanderen/nieuws/europese-ai-act-vanaf-2-augustus-gelden-nieuwe-transparantieregels-voor-ai    

2;     https://www.nieuwsblad.be/binnenland/belgie-klimt-naar-europese-top-voor-ai-gebruik-bij-bedrijven/156284682.html 

3.     https://publicaties.vlaanderen.be/view-file/82914 

4.     https://blog.atabix.nl/ai-act-fase-4-de-nieuwe-verplichtingen-en-hoe-je-hieraan-voldoet   

5.      https://digital-strategy.ec.europa.eu/nl/policies/regulatory-framework-ai

6.     https://economie.fgov.be/nl/themas/online/artificiele-intelligentie/veelgestelde-vragen-over