Note Notes 2026
EU AI Office, European AI Board, national authorities: who does what
Translated from the Italian original.
In June 2026 the EU AI Office published its first annual activity report, after becoming operational in February 2025. The figures indicate the workload: over 200 assessment procedures on GPAI models, 14 operational guidelines published, 7 codes of practice promoted or supported. In parallel, in Italy, AgID and the Garante Privacy (the Italian Data Protection Authority) are defining competences and coordination procedures. For Italian companies looking for operational help desks on AI Act questions, understanding who does what is a prerequisite for getting concrete answers instead of institutional buck-passing.
The point is that governance is multi-level and the right contacts depend on the question.
EU AI Office: the new central authority
The AI Office was formally established by a European Commission decision of February 2024 and began operating in the following months. Organizationally it sits within DG CNECT (Directorate-General for Communications Networks, Content and Technology) of the European Commission. It has around 140 staff, growing, organized in five units: regulation and compliance, AI safety, excellence in AI, AI for societal good, AI innovation and policy coordination.
The AI Office’s direct competences are substantial. Supervision of GPAI models with systemic risk (notification, assessment, mitigation). Development of the codes of practice encouraged by Article 56 of the AI Act. Support to the Commission in developing delegated and implementing acts. Coordination with national authorities and international bodies (OECD, G7, ISO). Point of contact for GPAI providers and for third-country authorities.
Indirect competences include monitoring the overall implementation of the AI Act in the Member States, supporting the development of harmonized standards, publishing interpretative guidelines. For an Italian company with doubts about a complex interpretation of the AI Act, the AI Office is the final reference point for EU interpretation, although it does not replace the national competent authorities for inspections.
European AI Board: the representation of the Member States
The European AI Board, officially the AI Board in the text of the Regulation, is composed of representatives of each EU Member State. It has an advisory function toward the Commission and the AI Office, contributes to the consistent application of the AI Act across Member States, and facilitates the exchange of experience between national authorities.
To frame the enterprise context of application, the summary of the Stanford AI Index 2026 is useful. For Italy, the representative on the Board is a government appointee, currently part of the delegation that MIMIT coordinates on AI matters at European level. Board meetings are monthly, decisions are mainly of a guidance nature. The Board is not an inspection authority, it does not impose penalties, it does not issue binding opinions on individual cases. It is a coordination space, important but not operational.
Under the Board operate thematic subgroups dedicated to specific areas of the AI Act: market surveillance, GPAI, fundamental rights, scientific support. The subgroups include experts from national authorities and in some cases representatives of industry and civil society.
Italian national authorities: AgID and Garante Privacy in coordination
In Italy the institutional framework of the AI Act is being formally defined with the bill adapting national law to the Regulation. The orientation consolidated in 2026 sees AgID as the main competent authority for most AI Act matters, with the Garante Privacy retaining specific competence over personal data processing.
The concrete application of the AI Act to marketing is one of the most frequent grounds of engagement with AgID and the Garante. AgID (Agenzia per l’Italia Digitale, the Agency for Digital Italy) has competences over public administration digitalization, digital identity, national cloud infrastructure. Adding AI Act competences required structural reinforcement: a new AI compliance division activated in 2026 with 30-50 dedicated staff. AgID is the authority that will receive notifications of high-risk systems placed on the market, conduct inspections, and impose penalties within its remit.
The Garante Privacy retains competence over everything concerning personal data in AI systems. The overlap with AgID is managed through coordination protocols. For Italian companies the practical point is that some procedures may require dual communication (notification to AgID for the AI Act part, possible communication to the Garante for the privacy part). Operational protocols are simplifying this duplication.
The other authorities with sectoral competences
The AI Act provides that existing sectoral authorities retain competence in their areas of specialization. Banca d’Italia on AI applied to credit and banking services. CONSOB on AI in financial markets. AGCOM on AI in communication and media services. AGCM on AI from a competition standpoint. ENAC on AI in aviation. AIFA on AI in medical devices and pharmaceuticals.
For an Italian company operating in regulated sectors, the framework of competent authorities is therefore multidimensional. A fintech company may deal with Banca d’Italia (sectoral regulation), AgID (AI Act), the Garante Privacy (personal data), CONSOB if listed. Compliance requires internal coordination among these relationships.
Sectoral authorities are developing AI guidelines specific to their fields. In December 2025 Banca d’Italia published guidance on the use of AI in credit scoring. CONSOB is working on guidelines on robo-advisors and on the use of AI in market controls. AGCOM has issued guidance on AI transparency in media content.
Help desks: whom to contact for what
A concrete question that often comes from Italian companies: “I have to notify an AI system, whom do I contact?” The answer has several parts. To notify a high-risk AI system placed on the market: AgID, through the dedicated portal under implementation (in the meantime via certified email, PEC, to the official mailbox). To notify a serious incident: AgID, with possible involvement of the relevant sectoral authority.
For a preliminary opinion on a system under development: the AI Act provides for regulatory sandbox mechanisms that in Italy will be operational from 2026. AgID will coordinate the process. For interpretative clarifications on specific rules: first your own legal counsel, then the EU AI Office guidelines, finally a formal request to AgID or to the relevant sectoral authority.
For complaints by parties harmed by AI systems: it depends on the nature of the harm. Discrimination: UNAR and the civil courts. Privacy violation: Garante Privacy. Product safety: the competent Ministry or the market surveillance authority. Fraud or crime: law enforcement. The single AI Act help desk will handle mixed cases, but the procedure has not yet been published in detail.
The real timelines of interactions with the authorities
In the first Italian cases, timelines vary. A high-risk system notification to the AgID portal should receive confirmation within 30 days; in practice the first cases took 60-90 days to close. An informal preliminary opinion can take 90-180 days, a formal one can exceed 12 months. A regulatory sandbox typically lasts 12-24 months.
For Italian companies the practical consequence is to plan ahead. The timelines of engagement with the authorities cannot be absorbed within a single quarter. If the goal is to launch a high-risk system for 2027, interactions with AgID should begin by the end of 2026. For GPAI providers, interactions with the AI Office have similar timelines.
The more structured companies are building ongoing relationships with the authorities through trade associations, thematic working groups, participation in public consultations. It is an approach that works better than sporadic case-by-case contact.
The opportunity of the early period
The first two to three years of the AI Act’s operation are a window of opportunity for proactive Italian companies. The authorities are building their practice, are open to dialogue with companies that present themselves constructively, and welcome input and suggestions to refine procedures. Those who position themselves as serious counterparts in 2026-2027 build relationships that will count in the coming years.
Trade associations are playing a substantial role. Confindustria, AssoSoftware, Anitec-Assinform, AICA and other representative bodies are formal counterparts of the authorities. For Italian SMEs, participating through these representatives is an efficient way to be heard. AIPIA, the Italian association of AI professionals, is also a space where individual professionals contribute to defining practices. Find out more with AIPIA about the technical working groups underway on the Italian implementation of the AI Act.
How to prepare for dealing with the authorities
Four principles that work in practice. Transparency: approach the authorities with clear and truthful documentation, even where there are weak points. Authorities appreciate transparency; they will discover the weak points anyway if concealed. Timing: engage ahead of problems, do not wait for inspections. Authorities are more cooperative with those who show up at the design stage than with those who arrive when the problem has already exploded.
Technical competence: bring to the discussions people who know what they are talking about, not only lawyers or commercial consultants. AI authorities have technical staff who assess substance, not just form. Continuity: maintain a structured relationship over time, not only in moments of crisis. Companies the authorities know as regular counterparts start from a better position of operational trust.
The AI institutional system under construction is imperfect, evolving, occasionally slow. It is, however, the space within which European AI innovation will be defined in the coming years. Understanding it, frequenting it, helping to shape it is a strategic investment, not an ancillary cost. The Italian companies that do so will be the ones that find room in the European market now taking shape.