EU AI Act in Estonia: Who Enforces It and What to Do Now
Estonia is set to name the TTJA as AI Act authority, with the AKI on data. Who enforces it, the penalties, and the 2027 high-risk timeline.
You do not wait for an Estonian law before the EU AI Act applies to you. Regulation (EU) 2024/1689 is directly applicable in Estonia from the dates set in the Regulation itself — Article 5's prohibited practices have applied since 2 February 2025, and the obligations bite regardless of where Estonia is in finalising its national enforcement set-up. For e-Estonia's deployer-heavy economy — a country that runs more public services through software than almost anywhere in the EU — the practical question is rarely "is this transposed yet?" It is "which of my AI systems are caught, what must I do, and who in Tallinn will ask about it?"
This guide sets out the competent authorities Estonia is designating, what is still being finalised, the corrected timeline after the Digital Omnibus, the penalties, and a concrete worked example for an Estonian SME.
A regulation, not a directive — no Estonian transposition needed
The EU AI Act is a Regulation. By virtue of Article 288 of the Treaty on the Functioning of the EU, it takes effect in Estonia directly, on the Regulation's own calendar, with no Estonian implementing statute required to make its obligations binding. This is the single most important point for Estonian companies, and the one most often misunderstood: the absence of a dedicated national act does not buy you time.
As of June 2026 there is no confirmed standalone Estonian AI Act implementing law. What a national measure does — when it arrives — is plumb the Regulation into Estonian administrative law: confirm which body holds which role, set domestic procedure for fines, and align existing sectoral supervision. It does not create or postpone the substantive duties. Those are already running.
So the prohibitions in Article 5 are live in Estonia today. The General-Purpose AI (GPAI) obligations in Chapter V have applied since 2 August 2025. And the high-risk regime arrives on its own schedule (below) whether or not an Estonian law has been passed by then.
Who enforces the EU AI Act in Estonia?
Estonia has adopted a multi-authority supervision model — there is no single AI "super-regulator." Responsibility is shared between the body lined up for AI systems, the data-protection supervisor, and, for general-purpose models, an EU-level body in Brussels.
TTJA — the body earmarked for AI systems
The TTJA — the Consumer Protection and Technical Regulatory Authority (Tarbijakaitse ja Tehnilise Järelevalve Amet) — is set to act as Estonia's competent authority for AI systems, and has already begun issuing public guidance for businesses. By its own account, the TTJA describes the competent-authority role as one it will fulfil going forward, once the designation is settled. It is Estonia's established market-surveillance and technical-regulation body, so AI Act market surveillance sits naturally alongside its existing product-safety and conformity remit.
One thing to state plainly rather than overstate. The Regulation requires each Member State to designate, under Article 70, both a market-surveillance authority and a single point of contact for the EU AI Office, plus a notifying authority for conformity-assessment bodies. The TTJA has been earmarked as the competent authority for AI systems — it is the body that will supervise AI systems and already publishes guidance — but the full single-point-of-contact and notifying-authority designation is still being finalised. Treat the TTJA as your enforcement counterpart for AI systems, and watch for confirmation of the remaining Article 70 roles rather than assuming a single body already holds all of them.
A note on timing: the Article 70 designation deadline was 2 August 2025, and a number of Member States — Estonia among those still finalising — did not complete the full designation by that date. A delay in naming authorities does not delay the obligations; it only affects the machinery that enforces them.
AKI — data protection in AI processing
The AKI — the Data Protection Inspectorate (Andmekaitse Inspektsioon) — supervises the data-protection dimension of AI. Where an AI system processes personal data — which covers almost every high-risk use in Annex III — the GDPR and Estonia's Personal Data Protection Act (PDPA) continue to apply in full, supervised by AKI, on top of the AI Act duties the TTJA is set to supervise.
In practice this means two authorities can have a legitimate interest in the same system. An Estonian company running an AI tool that screens job applicants or scores credit answers to the TTJA on AI Act market surveillance and to AKI on lawful processing, automated-decision safeguards (GDPR Article 22), and data-protection impact assessments (GDPR Article 35). The documentation built for one frequently informs the other.
The EU AI Office — general-purpose AI models
Estonian companies that develop and place general-purpose AI models on the market are supervised not by the TTJA but directly by the EU AI Office in Brussels, which has competence over the GPAI obligations in Articles 53 and 55. Fines on GPAI model providers are imposed by the Commission under Article 101, separately from the national penalty route. Most Estonian startups build on top of third-party models rather than shipping their own foundation models — if that is you, this layer is not your primary concern, but know where the line sits.
Estonia's national AI strategy: the "Kratt" plan
Estonia's AI policy is anchored in the AI and Data Action Plan 2024–2026, known by the nickname "Kratt" — a creature from Estonian folklore, a servant assembled from odds and ends and brought to life to do its owner's work, long adopted as the country's mascot for applied AI. The plan is led by the Ministry of Economic Affairs and Communications (MKM) and carries a budget of roughly €85 million. It sits alongside the broader Data and AI White Paper 2024–2030, which sets the longer strategic direction.
For a business, the relevance is context, not obligation: the "Kratt" plan signals that Estonia is investing in adoption and public-sector AI rather than building a punitive enforcement apparatus. It does not change a single duty under the Regulation. Useful background; not a compliance input.
A regulatory sandbox — expected, not yet confirmed
The Regulation requires every Member State to establish an AI regulatory sandbox by 2 August 2026 (Articles 57–59), with priority and free access for SMEs and start-ups (Article 58). Given Estonia's startup density and the MKM's posture, a national sandbox is widely expected — but as of June 2026 the operational details are not confirmed. If you are an Estonian SME planning a high-risk product, treat a sandbox as a likely future route to supervised testing, and check the MKM and TTJA channels for the launch rather than relying on it now.
How the EU AI Act classifies your systems
The Regulation does not classify companies or tools — it classifies uses. Four tiers, plus a separate cross-cutting category for general-purpose models:
| Tier | What it covers | What it means for you |
|---|---|---|
| Prohibited (Article 5) | Social scoring, untargeted facial-image scraping, manipulation exploiting vulnerabilities, most real-time public biometric identification, workplace/education emotion recognition | Banned outright — live since 2 February 2025 |
| High-risk (Article 6 + Annex III) | Credit scoring, recruitment and worker management, biometrics, essential services, and five other areas | Full obligations: risk management, documentation, human oversight, conformity assessment |
| Limited-risk (Article 50) | Chatbots, deepfakes, synthetic-content and AI-interaction disclosure | Transparency duties — tell people they are dealing with AI |
| Minimal | Everything else (spam filters, most analytics) | No mandatory AI Act obligations |
| GPAI (Chapter V) | General-purpose models themselves | Separate model-provider duties (Articles 53, 55), supervised by the AI Office |
Note that General-Purpose AI is a cross-cutting category, not a fifth risk tier. The system you build on top of a general-purpose model is classified by its own use — an AI chatbot for customer service is limited-risk under Article 50; the same model wired into a credit decision is high-risk. Classify by what the system does, never by the brand of model underneath it. For where the deployer/provider line falls, see the provider vs deployer guide.
The EU AI Act timeline as it applies in Estonia
| Date | What applies |
|---|---|
| 2 February 2025 | Article 5 prohibited practices and Article 4 AI literacy — in force, enforceable now |
| 2 August 2025 | GPAI obligations (Chapter V), governance and the AI Office, Article 99 penalties |
| 2 August 2026 | General application including most Article 50 limited-risk transparency duties |
| 2 December 2026 | CSAM / "nudifier" ban and Article 50 content-marking / watermarking — adopted (Digital Omnibus) |
| 2 December 2027 | Stand-alone high-risk AI systems (Annex III) — adopted (Digital Omnibus) |
| 2 August 2028 | High-risk AI as safety components of Annex I regulated products — adopted (Digital Omnibus) |
The 2026–2027 dates shown above exist under the Digital Omnibus — the Commission's amendment package — which deferred the high-risk regime to fixed calendar dates: 2 December 2027 for stand-alone Annex III systems, 2 August 2028 for Annex I product-embedded ones, and introduced the 2 December 2026 CSAM-ban and content-marking deadline. This package is now adopted. The European Parliament approved it on 16 June 2026 and the Council formally adopted it on 29 June 2026, with entry into force on Official Journal publication, expected before 2 August 2026. Stand-alone high-risk Annex III systems now apply from 2 December 2027. The new dates are fixed calendar dates — not tied to the availability of harmonised standards, a "stop the clock" version that was proposed and rejected.
And note what did not move: GPAI, the Article 5 prohibitions, Article 4 literacy, and most Article 50 transparency duties keep their original dates.
Penalties: what an Estonian company actually faces
Penalties sit in Article 99 of the Regulation and will be applied through Estonia's domestic enforcement procedure. There are exactly three tiers, each "whichever is higher" of a fixed sum or a percentage of total worldwide annual turnover:
| Breach | Maximum fine | Paragraph |
|---|---|---|
| Article 5 prohibited practices | €35,000,000 or 7% | Article 99(3) |
| Most other obligations — high-risk requirements, provider (Art 16) and deployer (Art 26) duties, Article 50 transparency | €15,000,000 or 3% | Article 99(4) |
| Incorrect, incomplete, or misleading information to authorities or notified bodies | €7,500,000 or 1% | Article 99(5) |
Two points matter most for Estonia's SMEs. First, the third tier is 1%, not 1.5% — a figure the market literature often gets wrong; there is no 1.5% tier anywhere in the Regulation. Second, and more reassuring: for SMEs and start-ups, Article 99(6) caps the fine at the lower of the fixed amount or the percentage. For a small Estonian company, the percentage almost always governs — and turnover-based exposure is a fraction of the headline figures. GPAI-model fines are the separate Article 101 instrument (up to €15M or 3%, imposed by the Commission).
Worked example: Tallinn Tasku OÜ
Tallinn Tasku OÜ is a fictional ~70-employee fintech-SaaS company in Tallinn, roughly €9 million turnover. It deploys two third-party AI systems: a credit-scoring model it licenses to assess loan applicants, and an AI CV-screener for its own hiring.
Role. Tasku did not build either system — it uses them as licensed. That makes it a deployer under Article 26, not a provider. Most Estonian SMEs land here, and the deployer duties are lighter than founders fear.
Classification. Both uses are high-risk under Annex III: creditworthiness assessment falls under Annex III point 5(b), and recruitment/worker-selection under point 4. So Tasku carries Article 26 deployer obligations on both — human oversight, monitoring, using the system per its instructions, and keeping logs.
FRIA. Because Tasku deploys a creditworthiness-scoring system (Annex III 5(b)), it is within the narrow set of deployers that must complete a Fundamental Rights Impact Assessment (FRIA) under Article 27 — distinct from the GDPR Article 35 DPIA, though the two can share a factual base. The CV-screener, as a private-employer deployment, does not by itself trigger the Article 27 FRIA.
Exposure. Suppose Tasku breaches a high-risk deployer obligation. The headline is "€15M or 3%," but Article 99(6) caps an SME at the lower figure: 3% of €9M is about €270,000 — real, but two orders of magnitude below €15 million. Right-sizing the fear is part of getting the response proportionate.
The provider trap. If Tasku starts fine-tuning the credit model on its own data, it risks crossing the Article 25 line — a substantial modification can flip a deployer into a provider, pulling in the full Article 16 provider stack and a fresh conformity assessment. Many Estonian engineering teams do not know this threshold exists.
Who asks. Tasku reports and answers to the TTJA on AI Act market surveillance. Because both systems process personal data at scale, AKI also has a legitimate interest in the lawfulness of the processing and the automated-decision safeguards.
How Confir helps
Confir is an EU-hosted compliance tool for exactly this work. Its classification engine is rule-based and deterministic — it encodes Articles 5 and 6 with Annex III logic as explicit rules, so the same intake always produces the same finding with a human-readable explanation of which rule fired. For a company like Tasku, it registers each AI system, derives the deployer-versus-provider role, runs the Article 27 FRIA where it is in scope, and generates the Annex IV technical-documentation pack and the Article 47 / Annex V Declaration of Conformity. Start with an AI use-case register to see what you actually hold.
What Estonian companies should do now
Today (Article 5 is live): Audit any AI use touching social scoring, untargeted facial-image scraping, manipulation of vulnerable people, or workplace/education emotion recognition. If a use fits and no exemption applies, stop or restructure it — the prohibition has been enforceable since 2 February 2025.
Before 2 August 2026 (Article 50): Make sure any chatbot, voice assistant, or AI-generated-content feature tells users they are dealing with AI and marks synthetic content. This is the near-term date that is not deferred.
Through 2026–2027 (high-risk prep): Build your AI inventory and classification, apply the Article 6(3) filter to confirm which systems are genuinely high-risk, assign provider/deployer roles, and start documentation. Build against 2 December 2027, the now-adopted high-risk deadline.
Ongoing: Watch for confirmation of Estonia's full Article 70 designation — the formal naming of the TTJA as competent authority, single point of contact, and notifying authority — and for the regulatory sandbox launch. Follow the TTJA's published guidance — its first enforcement signals will tell you which sectors and obligations it is prioritising.
Related guides
- Provider vs deployer: which role are you?
- Article 26 deployer obligations
- Article 27 and the FRIA requirement
- What a Fundamental Rights Impact Assessment is
- AI use-case register
- EU AI Act in Germany
- EU AI Act in Ireland
- Fintech AI compliance
Manage your EU AI Act compliance in one place
Confir automates risk classification, technical documentation, and audit trails for any company. No consultants. No 6-month projects. 14-day free trial.
Start free trial →