Germany passed its comprehensive implementing law only four days before the deadline, combining a new central coordination centre (KoKIVO) at the Bundesnetzagentur with continued sectoral oversight by BaFin, BfArM, BSI and the data protection authorities.
Same regulation, eleven different realities.
The EU AI Act sets the rules, but which authority enforces them, what national penalties apply, and whether a regulatory sandbox actually exists are all determined at member state level, not by Brussels. The table was compiled on 4 August 2026. Since then the Digital Omnibus has changed the frame: the high-risk rules now apply from 2 December 2027 (Annex III) and 2 August 2028 (Annex I), and a national sandbox must be operational by 2 August 2027 instead of 2 August 2026. Where a regulator has published something since August, the country note carries a dated update with its official source. Rows without an update were last checked on 4 August 2026. Several data points remain explicitly unclear rather than guessed.
11 countries · as of 2026-11-17
- Spain is furthest along in practice: AESIA has been operational and sanctioning since 2025, and Spain closed the first completed EU AI Act regulatory sandbox on 15 June 2026, even though its underlying organic law is still in parliament.
- Germany passed its implementing act, the KI-MIG, only four days before the deadline, entering into force 29 July 2026 and designating the Bundesnetzagentur as central market surveillance authority.
- Belgium and Sweden have no enacted national implementing law as of 4 August 2026; Belgium has missed the authority-designation deadline by a full year with enforcement possible only through the courts.
- Denmark was the first EU member state to legislate, with its authority designated back in April 2024, while Poland signed its law on 24 July 2026 but its new supervisory commission, KRiBSI, will not be fully staffed until November 2026.
- Austria has no dedicated implementing act at all and, per the last verified check in April 2026, still lacked a formally designated market surveillance authority, relying only on the RTR advisory service centre.
Sorted by how far the national implementing act has got. Where the position is unclear, the table says so rather than guessing.
| Country | Implementing act | Authority | Sandbox | Penalties |
|---|---|---|---|---|
| Germany DE | In force KI-MIG, in force since 29 July 2026 | Bundesnetzagentur (Federal Network Agency) | operational Pilot sandbox concluded March 2026; full Art. 57 readiness unconfirmed | AI Act maxima apply; exact KI-MIG fine procedure not fully verified |
| Italy IT | In force Legge 132/2025; Legislative Decree 179/2026 in force since 23 Oct 2026 | ACN, Bank of Italy, CONSOB, IVASS and the Garante, each for a defined scope | announced Joint AgID/ACN sandbox planned, access details not yet published | Up to EUR 35m/7% turnover planned, aligned with AI Act, via pending decree |
| Denmark DK | In force Law No. 467 (2025); revised version in force since 2 Aug 2026 | Digitaliseringsstyrelsen (Danish Agency for Digital Government) | operational Two sandbox rounds ran 2024-2025; not listed as ongoing April 2026 | Administrative fines and criminal penalties; exact amounts unconfirmed |
| Poland PL | Adopted AI systems law signed 24 July 2026; KRiBSI not yet operational | KRiBSI (Komisja Rozwoju i Bezpieczeństwa Sztucznej Inteligencji) under Ministry of Digitisation | announced Planned once KRiBSI is operational, targeted November 2026 | Administrative sanctions planned; exact fine amounts unclear |
| Ireland IE | Adopted Regulation of Artificial Intelligence Act 2026, signed 21 July 2026 | AI Office of Ireland (Department of Enterprise, Tourism and Employment) | announced Sandbox planned under Head 35; access details unconfirmed | AI Act maxima apply; CCPC gains new administrative sanction powers |
| France FR | In parliament PJL DDADUE 3, passed Senate 19 Feb 2026, not yet enacted | Not yet designated (DGCCRF proposed as coordinator) | unknown CNIL sandbox exists but not confirmed as Art. 57 instrument | No national penalty provisions beyond AI Act framework identified |
| Spain ES | In parliament Organic Law bill in Congress since 28 May 2026, not yet passed | AESIA (Agencia Española de Supervisión de Inteligencia Artificial) | operational First EU sandbox completed 15 June 2026, new rounds announced | Up to EUR 35m/7% turnover proposed; AESIA already sanctioning since Aug 2025 |
| Netherlands NL | Draft Uitvoeringswet AI-verordening draft; consultation closed 1 June 2026 | Autoriteit Persoonsgegevens + RDI, coordinating ~10 sector regulators | announced Single-portal sandbox design proposed March 2025, not yet operational | Not yet set; implementing act to define enforcement powers |
| Sweden SE | Draft SOU 2025:101 inquiry; no bill confirmed passed as of 4 Aug 2026 | PTS (proposed coordinator) + IMY + Finansinspektionen; not yet law | unknown PTS-run sandbox planned from 2 Aug 2026, status unconfirmed | AI Act maxima proposed (EUR 35m/7%); not yet enacted |
| Austria AT | None No dedicated implementing act identified; RTR service centre only | Not yet designated | unknown No concrete Austrian Art. 57 sandbox programme found | Not yet set |
| Belgium BE | None No implementing act; draft bill only, designation deadline missed since Aug 2025 | Not yet designated (BIPT planned as lead regulator) | none No Belgian sandbox programme identified | Not yet set; enforcement possible only through the courts |
Italy is the only country here with a national AI framework law fully in force since 2025, but the implementing decrees that actually define sanctions, authority powers and sandbox access only received preliminary cabinet approval in June 2026 and were not confirmed published by 4 August 2026, with a legal deadline of 10 October 2026.
Update 2026-10-08 Legislative Decree 179/2026 was published on 8 October 2026. It designates ACN, the Bank of Italy, CONSOB, IVASS and the Garante as market surveillance authorities for defined scopes and creates the national AI regulatory sandbox. It entered into force on 23 October 2026. Gazzetta Ufficiale, D.Lgs. 179/2026
Denmark was the first EU member state to enact AI Act implementing legislation and had its authority designated back in April 2024, well ahead of most peers, though a later EU Parliament overview did not list its sandbox as currently running.
Update 2026-09-14 Digitaliseringsstyrelsen completed its first 20 inspections of prohibited AI practices: ten in telemarketing and ten in online casino and betting. It found no use in breach of Article 5. Digitaliseringsstyrelsen, 14 Sep 2026
Poland signed its AI law just before the deadline, but the new KRiBSI supervisory commission will only get a chair in October 2026 and reach full staffing in November 2026, leaving a real operational gap at the 2 August 2026 EU deadline.
Update 2026-10-28 KRiBSI held its first meeting on 28 October 2026. Electronic forms for individual opinions and complaints have been available since 29 October 2026. gov.pl, KRiBSI
Ireland moved fastest of the distributed-model countries, enacting its law and standing up the AI Office of Ireland with a named CEO within ten days, coordinating roughly 13-15 sectoral regulators including the CCPC.
Update 2026-10-14 The first Board of the AI Office of Ireland was appointed on 5 August 2026. In October 2026 Ireland announced an AI regulatory sandbox through the office. Department of Enterprise, Oct 2026
France's proposed model splits oversight across roughly fifteen sectoral authorities coordinated by DGCCRF, with no single authority yet formally notified to the Commission, a structure French trade press has criticised as illegible for businesses.
Spain is the most operationally advanced country covered here: AESIA has run and sanctioned under a real mandate since 2025 and Spain closed Europe's first completed AI Act regulatory sandbox in June 2026, all while its foundational organic law is still working through Congress.
Update 2026-10-20 AESIA presented 16 practical guides from the Spanish AI regulatory sandbox. AESIA, 20 Oct 2026
The Netherlands has chosen a cooperative model spreading oversight across roughly ten existing sector regulators coordinated by AP and RDI, but as of 4 August 2026 the implementing law has not been submitted to parliament and remains only a closed consultation draft.
Update 2026-09-03 The RDI published its assessment of whether the draft Dutch implementation act can be executed and enforced. It sets out the tasks the RDI plans to take on, jointly with the AP. RDI, 3 Sep 2026
Sweden's entire AI Act framework still rests on an unpassed government inquiry, SOU 2025:101, with no source found confirming that implementing legislation actually cleared the Riksdag by the 2 August 2026 deadline, leaving the authority landscape unresolved.
As of the last verified check in April 2026, Austria's Chancellery could not give a date for formally designating a market surveillance authority, leaving the RTR AI service centre (a chatbot and advisory desk, not a formal Art. 70 authority) as the only visible national touchpoint; this should be re-verified before publication since sources only cover through April 2026.
Belgium missed the EU-wide 2 August 2025 deadline to formally designate its AI Act authorities and, per Belgian press reporting from 29 July 2026, still had no operational market surveillance authority a year later, leaving enforcement possible only through the courts.
The national layer is the work.
A tracker that treats the EU as one jurisdiction tells you the deadline. It does not tell you who inspects you, under which act, or whether the authority exists yet. We watch the national regulators alongside the EU sources and push changes as they land. Free for your own team, permanently.
Get free accessInformational content only, not legal advice. National positions change quickly and some entries are marked unclear on purpose.