The EU public administration is currently preparing its own AI systems for the implementation of the AI Act. The EDPS has an important role in this process, as the AI Act nominated it as the market surveillance authority for the AI systems of European Institutions, Agencies and Bodies (EUIs). A new EDPS Artificial Intelligence Unit was established in October 2024.
In this context, the EDPS launched, in cooperation with the AI Act Correspondents of the EUIs, a preliminary mapping exercise of all potential high-risk AI systems currently in use or planned by them; in one of the first mapping exercises launched by a market surveillance authority under the AI Act. The resulting report, the first document the EDPS publishes in its new role, demonstrates the diversity of AI systems already in the EUI ecosystem.
This voluntary mapping exercise had three main aims:
- First, to encourage EUIs to prepare early inventories and understand the AI systems they are using.
- Second, to ensure that the specificities of the EU public administration are identified early in the process so that implementing rules and guidance for the AI Act can take them into account.
- Third, to help the EDPS prepare for its new roles as market surveillance authority and notified body.
The mapping report was an opportunity for the EDPS to get to know better the ‘market’ it will supervise, helping to identify and define future priorities and areas where guidance could be beneficial for EUIs. Through this collaborative approach, the EDPS encouraged the EU public administration to ensure that its AI systems are trustworthy and safe, before the provisions on high-risk AI systems enter into application.
In the report, you will find a high-level summary, our findings, and the EDPS AI Unit’s key takeaways from the exercise. These are instrumental for the future strategy and priorities of the EDPS as a market surveillance authority. Some of these takeaways include:
- None of the EUIs declared in-use or planned prohibited AI practices as defined in Article 5 of the AI Act.
- At this early stage, EUIs often face difficulties when defining an AI system as high-risk and applying the filter of Article 6(3), which exempts certain systems under narrow conditions.
- It is important to ‘de-mystify’ high-risk AI systems and address the fear of adopting systems that fall into high-risk areas. Such systems can also bring benefits and efficiency. To fall into the category of high risk rather means that these AI systems need to comply with certain legal obligations and safeguards.
- AI Systems in the Area of Freedom Security and Justice (AFSJ) are likely to be a future focus for the EDPS AI Unit.
- In the field of recruitment, AI systems offering job matching services are popular and their use is likely to grow.
This mapping exercise was carried out in the framework AI Act Correspondents Network (AIACN) - a platform for exchange for the EUIs, initiated by the EDPS after the AI Act entered into force in August 2024. The AIACN is a success story; initially an informal network, it was formally established in September 2024 and now meets twice a year.