The AI Act came into force a little more than a year ago, but the momentum continues to grow across the EU institutions bodies and agencies. On 7 October 2025, more than one hundred representatives gathered, both online and in Brussels, for the second meeting of the AI Act Correspondents Network.
The event highlighted the importance of a harmonised approach among EU Institutions, agencies and bodies (EUIs) as they work towards the implementation of the AI Act in the EU public sector. It also marked a significant milestone on the path towards the effective enforcement of AI regulations - promoting a responsible use of AI in public administration while ensuring that innovation goes hand in hand with the protection of fundamental rights.
I had the pleasure of opening the meeting by recalling the proactive efforts of EDPS in the field of AI governance within the EU institutional framework. Just one year ago, we established a dedicated AI Unit to prepare for our role as the competent authority for EUIs under the AI Act, with enforcement powers coming into full power in August 2026. I also stressed our dual objective of safeguarding fundamental rights, health, and safety, while at the same time supporting the use of AI technologies to make the EUIs more efficient and responsive to the needs of people.
The day’s discussions began with Professor Alessandro Mantelero from the Polytechnic University of Turin, who presented the methodology and structure of an in-depth analysis of the Fundamental Rights Impact Assessment (FRIA). His address reminded us that assessing the impact of AI on individuals and society is pivotal in ensuring AI applications respect and promote fundamental rights, at the same time preventing potential rights infringements.
From the European Commission's AI Office, Killian Gross shared updates on the practical implementation of the AI Act. He presented the work on the General Purpose AI guidelines and Code of Practice, as well as the 'Apply AI Strategy', elaborating on the high-risk AI systems guidelines and transparency guidelines, which are designed to facilitate compliance and protect public and institutional interests.
Representing the EDPS AI Unit, Sophia Hassel introduced the results of the EPDS' voluntary mapping of high-risk AI systems and prohibited practices within EUIs. The preliminary findings identified hundreds of high-risk AI systems under development or currently in use across EUIs. The survey found that most EU institutions use AI systems based on generative AI or machine learning, particularly in the areas of biometrics, HR, and migration/asylum management.
Colleagues from the Joint Research Centre (JRC), Ignacio Sanchez and Bertrand De Longueville, took the discussion further into practice, sharing insights into their scientific pilot projects such as GPT@EC and GPT@JRC. They provided a glimpse into their recent exploration of agentic AI, illustrating how these developments contribute to building safer and more trustworthy AI ecosystem.
In a crucial intervention, Rossella Andronico from the EDPS offered an insightful overview of the AI Act provisions that assign specific powers to authorities responsible for safeguarding fundamental rights. Her presentation shed light on the legal foundations and guiding principles underpinning the cooperation between, in particular, Market Surveillance Authorities and Fundamental Rights Authorities - an essential framework for ensuring effective oversight and regulatory compliance.
Concluding the event, Sonia Perez Romero, Head of the EDPS AI Unit, thanked all participants for their contributions and active engagement making the discussions even more fruitful. She also outlined the next major step in our shared journey: the upcoming EDPS AI Regulatory Sandbox pilot project. This initiative will explore practical ways to apply innovative regulatory approaches under Article 57(3) of the AI Act - a step that symbolises our ongoing commitment to learning by doing.
Looking back, it was encouraging to see how much progress we have made since the network’s first meeting last year. The AI Act Correspondents Network has become a living community of practice - a space where collaboration takes shape, challenges are openly discussed, and expertise is shared. With a common, forward-looking agenda for the continuous development of AI governance across the EU public sector, EU institutions are not only preparing for compliance but also helping to shape a European model of trustworthy, human-centric AI.
The next meeting of the network will take place on 10 February 2026. Until then, our collective efforts will continue - guided by the same spirit of cooperation and bringing together cross-institutional expertise.