AI Act national competent authorities and sandboxes: list by country

Every EU country has to appoint at least one market surveillance authority and one notifying authority to apply the AI Act, and name one market surveillance authority as its single point of contact. In the European Commission's list of single points of contact, which was last updated on 7 September 2026 and is revised continuously, names one for 21 of the 27 member states, but 12 of those designations still await final adoption at national level and six countries have no entry. Each country must also have at least one AI regulatory sandbox operational by 2 August 2027, a year later than first required.

The sandbox deadline moved with the Digital Omnibus on AI, Regulation (EU) 2026/1744, which entered into force on 27 July 2026, six days before the original date of 2 August 2026. Other changed dates are in our EU AI Act timeline.

The three roles every member state must fill

  • Market surveillance authority. Enforces the rules for AI systems, including the Article 5 bans and high-risk requirements. It can demand documentation and source code, order corrective measures, impose penalties set by national law and take complaints from anyone.
  • Notifying authority. Designates and monitors the notified bodies that carry out third-party conformity assessments of high-risk AI.
  • Single point of contact. One market surveillance authority that acts as the contact for the public, other member states and the EU.

Member states were expected to have these authorities in place by 2 August 2025, and had to publish how to reach them by that date (Article 70).

Cases handled by other supervisors

  • Financial institutions: the national financial supervisor, for high-risk AI used in direct connection with financial services (Article 74).
  • Biometric systems used for law enforcement, border management, justice or democracy, and high-risk systems in law enforcement, migration, justice and democratic processes: data protection authorities, or another authority meeting the Law Enforcement Directive's conditions for independent supervisors.
  • AI in regulated products such as machinery or medical devices: the existing product-sector supervisor, unless the country picks another authority.
  • EU institutions: the European Data Protection Supervisor.
  • AI systems built on a general-purpose AI model by the same provider, and AI integrated into very large online platforms or search engines: the Commission's AI Office, under the amended Article 75. For the first group, carve-outs keep areas such as regulated products, critical infrastructure and law enforcement national.

Single points of contact by country (Commission list, 7 September 2026)

Entries follow the Commission's list. The single point of contact is the first stop, not necessarily the supervisor for every sector.

Designation adopted (9 countries)

  • Denmark: Agency for Digital Government (Digitaliseringsstyrelsen)
  • Finland: Transport and Communications Agency (Traficom)
  • Greece: Hellenic Data Protection Authority
  • Ireland: AI Office of Ireland (Oifig IS na hÉireann)
  • Italy: National Cybersecurity Agency (ACN)
  • Latvia: Consumer Rights Protection Centre (PTAC)
  • Lithuania: Communications Regulatory Authority (RRT)
  • Malta: Malta Digital Innovation Authority (MDIA)
  • Poland: Commission for the Development and Security of Artificial Intelligence (KRiBSI)

Named, national decision still pending (12 countries)

  • Cyprus: Commissioner of Communications
  • Czech Republic: Czech Telecommunication Office (ČTÚ)
  • Estonia: Consumer Protection and Technical Regulatory Authority (TTJA)
  • France: Directorate General for Competition Policy, Consumer Affairs and Fraud Control (DGCCRF)
  • Germany: Federal Network Agency (Bundesnetzagentur)
  • Luxembourg: National Commission for Data Protection (CNPD)
  • Netherlands: Dutch Authority for Digital Infrastructure (RDI)
  • Portugal: ANACOM, the communications regulator
  • Romania: ANCOM, the communications regulator
  • Slovenia: Agency for Communication Networks and Services (AKOS)
  • Spain: Spanish Agency for the Supervision of Artificial Intelligence (AESIA)
  • Sweden: Swedish Post and Telecom Authority (PTS)

No single point of contact listed (6 countries)

Austria, Belgium, Bulgaria, Croatia, Hungary and Slovakia.

How the larger markets are organised

Germany

The federal government's draft AI market surveillance and innovation act (KI-MIG) gives the Bundesnetzagentur a central role, including the national sandbox (KI-Reallabor in German). The agency's guidance lists five notifying authorities under the KI-MIG, including the Federal Office for Information Security (BSI) for high-risk AI under point 1 of Annex III (biometrics). With Hesse's digital ministry and the federal data protection commissioner, the agency simulated a sandbox from May 2025 on two start-up use cases in medical AI and published the final report in March 2026.

Italy

Law No. 132 of 23 September 2025, in force since 10 October 2025, names two national AI authorities: the Agency for Digital Italy (AgID) as notifying authority, and the National Cybersecurity Agency (ACN) as market surveillance authority, with inspection and sanctioning powers, and single point of contact. The Bank of Italy, CONSOB and IVASS supervise AI used by financial institutions. AgID and ACN are tasked with setting up and jointly running sandboxes.

Spain

Spain ran a pilot sandbox under Royal Decree 817/2023 of 8 November 2023, adopted while the AI Act was still a proposal. AESIA describes it as Europe's first regulated AI testing environment and says it has concluded. It produced 16 practical guides on requirements such as risk management, human oversight and technical documentation, which AESIA plans to update for the Omnibus.

Netherlands

The RDI and the Dutch Data Protection Authority (AP) coordinate AI oversight, while sector regulators supervise AI in their own fields. Under the draft implementing act, as described in the RDI's September 2026 enforceability test, the RDI is also responsible for running the regulatory sandboxes. A 2025 proposal sets out a single multi-sector sandbox in which all relevant supervisors take part, drawing on a pilot that handled 40 questions from 25 organisations.

Ireland

The Regulation of Artificial Intelligence Act 2026, signed into law on 21 July 2026, set up the AI Office of Ireland as an independent statutory body that coordinates enforcement. Supervision is spread across 13 authorities designated so far, including the Central Bank, the Data Protection Commission, ComReg and the Workplace Relations Commission. The Irish sandbox is still under development.

Poland

By late July 2026 the President had signed the Act on AI systems. It creates KRiBSI, an independent collegial body with representatives of the competition, financial, broadcasting and electronic communications regulators, which will run regulatory sandboxes that micro, small and medium-sized firms can join free of charge. The digital ministry projected that it would start work in November 2026.

Denmark, Finland and Lithuania

  • Denmark: the agency's site says the Agency for Digital Government, the Danish Data Protection Agency and the Danish Court Administration are tentatively set to supervise the Article 5 bans, and that supervisors for high-risk AI are not yet settled. The agency and the data protection authority run a regulatory sandbox giving free guidance on GDPR and AI Act risk classification; the 2026 round's guidance period runs from September to December 2026.
  • Finland: Traficom coordinates a sandbox whose pilot phase runs from September 2026 to the end of April 2027. The full service starts on 3 May 2027.
  • Lithuania: the Communications Regulatory Authority is the single point of contact on the Commission's list, and the Innovation Agency operates the sandbox. General consultations are already available; the formal sandbox programme is planned for 2027.

What an AI regulatory sandbox offers under Article 57

A sandbox is a supervised, time-limited setting where a provider develops, tests and validates an AI system before market launch, under a plan agreed with the authority. Under the amended Article 57, countries may set one up jointly or join an existing sandbox that gives equivalent national coverage.

  • Cost: access is free for SMEs, including start-ups, although authorities may recover exceptional costs (Article 58).
  • Fines: participants who follow the sandbox plan and the authority's guidance in good faith face no administrative fines under the AI Act, but remain liable for damage to third parties.
  • Exit report: the authority provides an exit report and, on request, written proof of completed activities, which market surveillance authorities and notified bodies must take into account positively.
  • EU level: the Omnibus lets the AI Office run a Union-level sandbox for systems under its supervision, with priority for SMEs, start-ups and small mid-caps. The European Data Protection Supervisor may run one for EU institutions.
  • Common rules: the AI Office must publish a list of planned and existing sandboxes, and the Commission is to set detailed sandbox rules through implementing acts.

Testing in real-world conditions outside a sandbox

Under Article 60, providers of high-risk AI listed in Annex III and, since the Omnibus, of AI in products covered by Section A of Annex I can test in real-world conditions without a sandbox. The local market surveillance authority must approve a testing plan; silence for 30 days counts as approval only where national law provides for tacit approval. Tests must be registered, last no more than six months plus one six-month extension, and generally need participants' informed consent. A new Article 60a lets member states opt in to frameworks for testing AI in products under Section B of Annex I, such as vehicles, rail and aviation.

More national coverage is in our Capitals and AI Act sections.

Frequently asked questions

Who enforces the AI Act in each EU country?

National market surveillance authorities enforce the rules for AI systems, with one acting as single point of contact. The Commission's AI Office supervises general-purpose AI models and certain AI systems built on them or integrated into very large online platforms.

Which authority enforces the AI Act in Germany?

Under the draft KI-MIG, the Bundesnetzagentur is set to be Germany's main AI market surveillance authority and single point of contact, with sector regulators keeping some areas. The Commission's September 2026 list still marked the designation as pending final adoption.

What is the deadline for national AI regulatory sandboxes?

Each member state must have at least one AI regulatory sandbox operational by 2 August 2027. The original deadline was 2 August 2026, but Regulation (EU) 2026/1744 pushed it back by a year.

Is Spain's AESIA sandbox still running?

The pilot has ended. According to AESIA, the sandbox set up under Royal Decree 817/2023 has concluded, and its results were published as 16 practical guides on AI Act requirements.

Which EU countries have an AI regulatory sandbox?

Several have started: Spain and Germany have completed pilots, Denmark runs a sandbox offering regulatory guidance (current round ends in December 2026), and Finland is piloting one until the end of April 2027. Every member state must have at least one operational by 2 August 2027.