On 19 November 2026, ERR News reported that the European Commission has criticised Estonia for delays in applying the EU Artificial Intelligence Act nationally and wants an oversight framework set up quickly. Justice and Digital Affairs Minister Liisa Pakosta confirmed that changes are underway. She said decisions could come at the start of the next Riigikogu's term, and that she believes the EU should have a central AI oversight authority. ERR notes that the Act's oversight and enforcement provisions took effect in early August this year.
The strongest case for the Commission's impatience
The Commission's position is easy to defend. A rulebook with no regulator behind it protects no one. People harmed by a prohibited AI practice have nobody to complain to, and companies get no guidance. Compliant firms are also disadvantaged against rivals in states where enforcement is lax or absent.
The legal duty is clear. Article 70 of the AI Act says each Member State "shall establish or designate" at least one notifying authority and at least one market surveillance authority. Contact details for these bodies were due to be public by 2 August 2025, and states must give them adequate technical, financial and human resources (Article 70). The Commission's own page says it may open a formal infringement procedure against a state that fails to designate a market surveillance authority by that date. That page does not say whether any procedure has been opened (Commission page on market surveillance authorities).
Estonia is not alone. A tracker run by the Future of Life Institute, last updated 17 June 2026, counts only nine Member States as having designated both a market surveillance and a notifying authority. Twelve are partial, and six have designated nothing. Estonia is among the six, and the tracker says its market surveillance and notifying authorities "have not yet been appointed" (AI Act national implementation tracker). This is a third-party, work-in-progress tracker, so treat it as indicative. The Commission's page also shows Estonia's single point of contact as pending final adoption. Estonia is one of several laggards, but it is a laggard.
Why "centralise it in Brussels" cannot be the plan
Minister Pakosta's preference for EU-level oversight is a legitimate policy position, and a sensible one for a small state. A country of Estonia's size will struggle to staff a deep technical AI regulator, and fragmented national enforcement can produce 27 different readings of the same rule. A single European supervisor would arguably be better for innovators who operate across borders.
But that is an argument for changing the law, not for postponing compliance with it. The Commission's page says the AI Office supervises and enforces the rules for general-purpose AI models. National market surveillance authorities handle the rest: they can intervene when AI systems pose risks, access providers' documentation and data, request corrective measures and impose penalties. They must also accept complaints from any natural or legal person (Commission page). Unless the EU legislature changes that split, Estonia needs a national body however much it favours centralisation.
Pakosta told ERR that the delay was partly due to the EU amending its entire package of AI legislation in the meantime, and that Estonia is exploring the oversight question with other member states. Moving targets are a real excuse for a ministry drafting a statute. They are a weaker excuse for leaving businesses unsure who will supervise them, fifteen months after the Article 70 date.
What proportionate implementation looks like
The pro-innovation answer is the lightest structure that satisfies the Act, designated quickly.
- Reuse existing regulators. Article 70 lets states appoint any kind of public entity. Assigning AI Act duties to sector regulators that already know the markets they supervise avoids creating a new agency, with its own headcount and its own incentive to expand its remit.
- Designate before the whole statute is finished. A short designation act and a named single point of contact would give firms an address for questions. The detailed penalty regime and procedures can follow. Waiting for the next Riigikogu to do everything at once prolongs the uncertainty.
- Prioritise guidance over penalties. Estonia's digital-government reputation rests on services that work. The same attitude should apply here: clear public guidance, a visible helpdesk, and enforcement that targets the prohibited practices the Act actually bans, not routine software.
- Keep pressing for EU-level simplification. Estonia's push for stronger central coordination, which Pakosta says it is leading alongside Finland, is a reasonable use of its influence. It should run alongside national designation, not substitute for it.
There is also a plain cost to doing nothing. Estonian start-ups and public-sector buyers of AI tools cannot tell which body will interpret the rules or where to take a dispute. Legal uncertainty falls hardest on small firms, which cannot afford an in-house compliance team. A visible regulator with narrow, published priorities is better for them than a vacuum that fills with guesswork and overcautious compliance.
The bottom line
The Commission is right on the obligation and the minister is right on the long-term design. Only one of them is a matter of legal duty today. Estonia should use the existing legal framework to name its authorities, publish a contact point and say how it will proportionately enforce the rules. It can keep arguing in Brussels for a more centralised model. If the EU later moves oversight upward, a small, lean national authority will be easy to fold in. A late and sprawling one would be harder to unwind.