What EU AI Act enforcement actually looks like today
Enforcement powers went live on 2 August 2026. Most coverage since has focused on whether anyone has been fined yet. That is the wrong question, and answering it correctly is more uncomfortable than the headlines.
80 days to 2 December 2026 — two new Article 5 prohibitions, and Article 50(2) marking for generative systems already on the market. Full timeline →
Has anyone been fined under the EU AI Act yet?
As at 13 September 2026 we have found no confirmed, primary-sourced enforcement decision imposing a fine under the AI Act. The European Commission's own enforcement page, last updated 24 August 2026, describes the AI Office's powers and timeline but does not report any penalty decisions.
Reg. (EU) 2024/1689 as amended by Reg. (EU) 2026/1744 · more questions on this page
Does this reach your organisation?
- Yes if you place an AI system on the EU market, or put one into service in the EU, wherever you are established.
- Yes if you are established in the EU and use an AI system under your own authority.
- Yes if you are outside the EU and the output your system produces is used in the Union.
No EU entity and no EU customers of record is not an exemption. Article 2 turns on market placement and on where output is used. Check your position in 12 questions →
First, a correction to something you have probably read
Several widely-shared articles report specific first fines under the AI Act, with figures in the tens of millions and named sectors. We have not been able to verify a single one against a primary source.
The European Commission’s own enforcement page, last updated 24 August 2026, sets out the AI Office’s powers and timeline in detail. It reports no penalty decisions. Had tens of millions in fines been issued in the first week of August, that page would say so.
We are not repeating those figures, and you should be careful about citing them in a board paper. But the absence of a published fine is not the reassurance it appears to be, for the reason set out below.
What the AI Office can do today, with no new legislation required
These powers have applied since 2 August 2026. Each is drawn from the Commission’s published enforcement framework.
| Power | What it means in practice |
|---|---|
| Requests for information | Issued informally by the AI Office or formally by Commission decision. You must answer, and the answer itself is penalty-bearing. |
| Model evaluations | The AI Office, or independent experts it appoints, can require you to grant access to your model for evaluation. |
| Inspections of premises | Not a document request. An inspection of the provider’s premises. |
| Interviews | Of any person who may hold relevant information and consents. That includes your staff. |
| Restriction of availability | The AI Office can require measures including restricting a model’s public availability. For a company whose product is the model, that is an existential remedy, not a fine. |
| Penalty decisions | Where an intentional or negligent breach is established, the Commission may adopt a decision imposing penalties. |
Where this usually goes next
Three situations account for most people reading this page. Each has a different answer.
A deal is blocked on an AI questionnaire
Legal will not sign until you can evidence how AI is governed. HumanAudit’s AI Trust Package is a fixed $3,500 over five business days: a public trust page, a pre-filled SIG Lite / CAIQ / SSPA Section K questionnaire bank, and your AI inventory and classification.
You need ISO/IEC 42001 documentation
23 clause-mapped AIMS documents with all 38 Annex A controls pre-populated, editable and yours to keep, from $199. Or score your gaps first: 18 questions, free, no signup to begin.
You are not sure what reaches you
A request for information from the AI Office asks which AI systems you run, what each is for, which role you hold, and what is documented. The reply itself is penalty-bearing. Twenty minutes covers the same ground while nothing turns on the answer.
This reference is published by HumanAudit Inc. Not a law firm, not an accredited certification body, not a registered auditor. We build documentation, your counsel interprets it, and an accredited body of your choosing certifies you. How this is funded →
Three channels are pointed at your organisation right now
This is the part almost nobody has registered, and it is the reason “no fines yet” is not reassuring. The AI Office does not have to find you. It has built three routes for other people to report you, and all three are live.
The complaint tool
Any natural or legal person can submit a complaint alleging an infringement by a provider supervised by the AI Office. Competitors. Customers. Campaigners. Journalists.
The whistleblower tool
Your own staff. Individuals professionally connected to a provider of an AI system or a GPAI model can securely report violations. Secure means you will not know who.
The downstream channel
Your customers. A provider building on your general-purpose model can complain about your compliance with Articles 53 to 55, directly.
The practical consequence
Expert analysis. Regulators with limited capacity and an open complaints channel do not audit at random. They act on what arrives. That inverts the usual risk model: your exposure is no longer a function of how likely you are to be selected, it is a function of how many people around you know something is wrong and have a frictionless way to say so.
The employee who raised a concern about your hiring model and was told it was fine now has a secure reporting channel operated by the European Commission. That was not true before 2 August 2026.
The exposure most organisations have not priced
Failure to comply with a request for information, or supplying incorrect, incomplete or misleading information, carries its own fine. Up to €7.5 million or 1% of worldwide annual turnover for AI systems; the higher maxima apply for GPAI models.
Read that carefully
You can be substantively compliant and still be fined for how you answered. An RFI arrives with a deadline, it lands on someone who has never seen one, and the reply is assembled from whatever is to hand. If the inventory is out of date, the classification was never documented, or the accuracy figure in the reply does not match the one in the technical file, the reply is incorrect information.
This is the single strongest argument for having the documentation in order before anyone asks, and it has nothing to do with whether you think you are high-risk.
The penalty tiers, stated correctly
Most published summaries get this wrong, usually by attaching the 7% figure to high-risk breaches. It does not apply there.
| Breach | Maximum | Source |
|---|---|---|
| Article 5 prohibited practices | €35M or 7%, whichever is higher | Art. 99(3) |
| Other operator obligations, including high-risk and Article 50 transparency | €15M or 3% | Art. 99(4) |
| Incorrect, incomplete or misleading information to authorities | €7.5M or 1% | Art. 99(5) |
| GPAI model providers, Commission-enforced | €15M or 3% | Art. 101 |
| SMEs and start-ups | The lower of the fixed sum and the percentage | Art. 99(6) |
If a vendor, adviser or article tells you high-risk breaches carry 7%, they are reading a summary rather than Article 99. Penalties in full →
What is enforceable today, and what is next
| Date | Status | What |
|---|---|---|
| 2 Feb 2025 | enforceable | Article 5 prohibitions · Article 4 AI literacy |
| 2 Aug 2025 | enforceable | GPAI model obligations. The Commission has confirmed that violations dating from this point are in scope |
| 2 Aug 2026 | enforceable | Enforcement powers activate. Article 50 transparency · Article 49 registration · national market surveillance |
| 2 Dec 2026 | 80 days | Two new Article 5 prohibitions (NCII, CSAM) · Article 50(2) marking for systems already on the market |
| 2 Dec 2027 | upcoming | Annex III high-risk obligations |
| 2 Aug 2028 | upcoming | Annex I embedded high-risk |
2 December 2026 is the one being missed
Article 50(2) marking for systems already on the market is an engineering change with a lead time, not a policy document. If you shipped a generative feature before 2 August 2026 and have not implemented machine-readable provenance marking, you have roughly eleven weeks, and the work sits with a product team that probably has not been told.
Status labels on this page
Verified fact: The AI Office's investigative and sanctioning powers, the three reporting channels, the Art. 99 and 101 penalty tiers, the RFI penalty exposure, and every date above, taken from the European Commission's published enforcement framework (last updated 24 August 2026) and the consolidated Regulation.
Expert analysis: The assessment of how an open complaints channel changes the risk model, and the observation about RFI reply quality.
Unsettled: Whether any penalty decision has been adopted. We found none in primary sources as at 13 September 2026, and we do not repeat unverified figures circulating elsewhere.
The question an RFI would ask, asked now instead
A twenty-minute walkthrough covers the same ground a request for information would: which AI systems you run, what each is for, which role you hold, what is documented and what is not. The difference is that nothing turns on the answer today.
Not sure where you sit?
The classifier maps your system against Articles 5, 6, 50 and Annex III. Twelve questions, no email.
Frequently asked
Has anyone been fined under the EU AI Act yet?
As at 13 September 2026 we have found no confirmed, primary-sourced enforcement decision imposing a fine under the AI Act. The European Commission's own enforcement page, last updated 24 August 2026, describes the AI Office's powers and timeline but does not report any penalty decisions.
Several widely-shared articles claim specific first fines totalling tens of millions of euros; we have not been able to verify any of them against a primary source and we do not repeat them. The absence of a published fine is not evidence of an absence of risk, because the investigative machinery is live and operating.
What enforcement powers does the AI Office actually have right now?
Since 2 August 2026 the AI Office can send requests for information, perform model evaluations, issue requests for access to models, require providers to take measures including restricting a model's public availability, conduct interviews of any consenting person with relevant information, and conduct inspections of provider premises. Where it establishes an intentional or negligent breach, the Commission may adopt a decision imposing penalties.
Can our own employees report us to the AI Office?
Yes. The AI Office operates an AI Act Whistleblower Tool through which individuals connected to providers of AI systems or general-purpose AI models in a professional capacity may securely report violations. It also operates a public AI Act Complaint Tool allowing any natural or legal person to submit a complaint about alleged infringements by providers supervised by the AI Office, and a separate complaints channel for downstream providers using general-purpose AI models.
Can we be fined for answering a regulator's question badly?
Yes, and this is a separate exposure from the underlying obligation. Failure to comply with a request for information, or providing incorrect, incomplete or misleading information, may lead to fines. For AI systems these reach up to 7.5 million euro or 1 percent of worldwide annual turnover, whichever is higher. For general-purpose AI models the higher maxima apply. An organisation that is substantively compliant can still be penalised for how it responds.
What is the next EU AI Act deadline?
2 December 2026. Two new Article 5 prohibitions covering the generation or manipulation of non-consensual intimate material and child sexual abuse material take effect, and Article 50(2) machine-readable marking becomes applicable to generative systems that were already on the market before 2 August 2026.
Obligations, article by article
- Art. 5 prohibitions
- Art. 4 AI literacy
- Art. 50 transparency
- Art. 9 risk management
- Art. 10 data governance
- Art. 11 / Annex IV
- Arts. 12–13 logging
- Art. 14 human oversight
- Art. 15 accuracy & security
- Art. 17 QMS
- Arts. 43–48 conformity
- Art. 49 registration
- Art. 57 sandboxes
- Open source
- Art. 72 monitoring
- Art. 73 incidents
- Arts. 51–56 GPAI
- Art. 99 penalties
- Compliance checklist
- FRIA template (Art. 27)
- When Annex III does not apply →