Skip to content

GuideUpdated 4 November 2026

The EU AI Act for SMEs: what it requires of you and from when

If your SME uses AI tools, two things already apply to you today: training the people who use them, and telling people when they are talking to an AI or looking at fake AI-made content. The heavy high-risk obligations, such as using AI to filter CVs or assess people's creditworthiness, were postponed to 2 December 2027 by the Digital Omnibus. In Spain, the law that sets out who imposes penalties is still in Congress.

Xora Intelligence10 min read · 10 sources
Editorial photograph illustrating: The EU AI Act for SMEs: what it requires of you and from when

You use an AI chat to write emails, you have a web assistant that answers at night, and you are thinking about a tool that sorts CVs. Does the European AI law affect you? Yes. But far less than some people have told you, and not all at once.

In this guide we explain what Regulation (EU) 2024/1689 says, the so-called EU AI Act (Reglamento Europeo de Inteligencia Artificial), from the point of view of an SME that uses AI or commissions a custom one. With the dates that apply today, after the change approved this summer.

Who does it apply to, and what is your role?

The Regulation shares out obligations according to your role. Two of them matter to you:

  • Provider. Whoever develops an AI system, or has one developed, and places it on the market or puts it into service under their own name or brand. OpenAI is the provider of ChatGPT. Careful: if you ask a company to build you an assistant and you launch it under your brand, the definition in Article 3 makes you the provider.
  • Deployer. Whoever uses an AI system under their own authority in a professional activity. If your team uses an AI chat or a program with AI to work, your company is a deployer.

Most SMEs are only deployers. And the Regulation asks very little of someone who only uses AI, except for high-risk uses.

What are the risk levels?

The Regulation does not regulate "AI" in general. It regulates uses, and sorts them into four levels:

Level What it includes What happens
Prohibited Manipulating people with subliminal techniques, social scoring of people, inferring your employees' emotions at work, among others You cannot use it. Full stop
High risk AI that decides or influences serious matters in people's lives: hiring, assessing workers, granting credit, education, life and health insurance Strict obligations for whoever makes it and whoever uses it
Transparency Chatbots, AI-generated content, deepfakes You have to tell people
Minimal Almost everything else: drafting texts, summarising, translating, spam filters, invoice analysis No specific obligations

According to the European Commission, most AI systems used in the EU today are at the minimal level.

From when does each part apply?

The Regulation did not arrive all at once. It applies in phases, and the Digital Omnibus on AI changed several dates. The Commission proposed it on 19 November 2025, Parliament and the Council approved it in June 2026 and it was published as Regulation (EU) 2026/1744. It has been in force since 27 July 2026. It is no longer a proposal: it is law.

This is the timeline that applies as of 25 September 2026:

Date What starts to apply Who it affects
1 August 2024 The Regulation enters into force. No obligations yet Nobody yet
2 February 2025 Prohibited practices (Art. 5) and AI literacy (Art. 4) All companies that use or build AI
2 August 2025 Rules for general-purpose AI models, governance and the fines regime Mainly the big model providers
27 July 2026 The Omnibus enters into force. Article 4 is rewritten All companies that use or build AI
2 August 2026 General application. Transparency obligations of Art. 50 Anyone who puts a chatbot in front of people or publishes deepfakes; generative AI providers
2 December 2026 New prohibitions: AI that generates intimate images without consent or child sexual abuse material. Marking of generated content for systems already on the market before 2 August 2026 Generative AI providers, and anyone who uses it for that
2 December 2027 High-risk obligations of Annex III: employment, credit, education, insurance, among others SMEs that use AI in recruitment, employee evaluation or customer creditworthiness
2 August 2028 High risk in products that are already regulated (Annex I): machinery, toys, medical devices Manufacturers of those products

If you have read that "the Regulation is fully applicable from August 2026", that is out of date for high risk: that part was delayed.

What applies to you today as an SME that uses AI?

Two things. Neither requires filing any paperwork.

Train the people who use AI (Article 4)

Article 4 requires providers and deployers to take measures to support the AI literacy of their staff and of anyone using the systems on their behalf. In plain terms: the people who use the tool should know what it does, where it gets things wrong and what they must not feed into it.

The Omnibus softened this article. The original version asked you to ensure "a sufficient level" of knowledge. The new one asks you to take measures and makes clear that it does not oblige you to guarantee a specific level in each person. In addition, the Commission and the Member States take on a bigger role in supporting SMEs.

What to do in practice:

  • Short training, adapted to what each person uses.
  • Written rules of use: which tools, with which accounts, which data must never go in.
  • A record of what you have done.

We walk you through it step by step in how to train your team in AI. Article 99, which sets the Regulation's fines, does not include a penalty of its own for Article 4.

Tell people when they are talking to an AI (Article 50)

The transparency obligations apply from 2 August 2026. These are the ones that may concern you:

  • Chatbots and assistants. People have to know they are talking to an AI system, unless it is obvious from the context. The notice has to be clear and arrive, at the latest, at the first interaction. The rule requires it of the provider, who must design the system that way. If you use someone else's chatbot, check it. If it was built for you and runs under your brand, that obligation is yours.
  • Deepfakes. If you use AI to generate or manipulate images, audio or video that resemble real people, places or events and could pass as authentic, you have to say they are artificial. For clearly creative or fictional works, it is enough to indicate it without spoiling the work.
  • Texts on matters of public interest. If you publish AI-generated texts to inform the public on matters of public interest, you have to say so. You do not need to if a person has reviewed them and someone takes editorial responsibility.
  • Emotion recognition or biometric categorisation. If you use a system like that, you have to inform the people exposed to it. Remember that inferring your employees' emotions at work is prohibited, except for medical or safety reasons.

Technically marking generated content, with a signal a machine can detect, is the job of the tool's provider, not yours.

Our assistants introduce themselves as AI from the first message and say how to reach a person. If you want to see how we build them, look at chatbots.

What if I use AI in human resources or to grant credit?

Here everything changes. Annex III of the Regulation classes as high risk, among others:

  • Systems for recruiting staff: placing targeted job ads, analysing and filtering applications, evaluating candidates.
  • Systems for deciding about workers: promotions, dismissals, allocating tasks based on their behaviour or personal traits, or monitoring and evaluating their performance.
  • Systems for assessing the creditworthiness of individuals or giving them a credit score. Detecting fraud does not count.
  • Systems for pricing or assessing risk in life and health insurance for individuals.

There is one exception: an Annex III system may not be high risk if it only performs a limited procedural task, prepares an assessment or improves work a person has already done. But if it profiles people, it is always high risk.

If you use one of these systems, from 2 December 2027 you will have to meet the obligations of Article 26:

  • Use it according to the provider's instructions.
  • Put a person with training and authority in charge of overseeing it.
  • Make sure the data you feed it is relevant, if you control it.
  • Monitor how it works and notify the provider and the authority if you see a risk or a serious incident.
  • Keep the logs the system generates for at least six months.
  • If you use it at work, tell the affected workers and their representatives beforehand.

If you assess the creditworthiness of individuals or price life and health insurance, you will also have to carry out a fundamental rights impact assessment before using it (Article 27).

Until that date arrives, the GDPR (RGPD) already applies to any system that processes candidate or customer data. You will find it in AI and data protection.

Does it affect me? Typical SME cases

What you do Level What applies to you From
Chatbot or agent that serves customers on the web or WhatsApp Transparency It must introduce itself as AI in the first message. Train whoever manages it 2 Aug 2026 (notice) · 2 Feb 2025 (training)
Writing emails, website texts or posts with AI Minimal Train whoever uses it. If the text informs on matters of public interest and nobody reviews it, give notice 2 Feb 2025
Analysing invoices, extracting data from delivery notes, reconciling payments Minimal Train whoever uses it 2 Feb 2025
Summarising meetings or internal documents Minimal Train whoever uses it 2 Feb 2025
Product images or creatives generated with AI Minimal or transparency If it looks like a real person, place or event and could pass as authentic, say it is artificial 2 Aug 2026
AI that filters or scores CVs High risk Art. 26 obligations and prior notice to staff 2 Dec 2027
AI that decides shifts or evaluates employee performance High risk Art. 26 obligations and prior notice to staff 2 Dec 2027
AI that decides whether you give financing to an individual customer High risk Art. 26 and the impact assessment of Art. 27 2 Dec 2027
Detecting employees' emotions by camera or voice Prohibited You cannot use it 2 Feb 2025

What fines are there and who do they apply to?

Article 99 sets the maximums. They are caps for the whole EU, not the fine you will be given:

Infringement Maximum fine
Using a prohibited practice (Art. 5) €35 million or 7% of worldwide annual turnover
Breaching high-risk obligations as a user (Art. 26) or transparency obligations (Art. 50) €15 million or 3% of worldwide annual turnover
Giving false or incomplete information to the authorities €7.5 million or 1% of worldwide annual turnover

For a large company, the higher of the two figures applies. For an SME, the lower. The Omnibus extended that rule to small mid-caps. The article also requires taking into account the economic viability of SMEs and the seriousness, duration and damage of each infringement.

Who enforces it in Spain, and what about the Spanish law?

The authority is the Spanish Agency for the Supervision of Artificial Intelligence (AESIA), created by Royal Decree 729/2023 and based in A Coruña.

The Regulation applies directly, but leaves each country to set the penalties regime and designate authorities. In Spain that will be done by the Organic Law for the good use and governance of artificial intelligence. This is where it stands:

  • 11 March 2025. The Council of Ministers approves the draft bill.
  • 26 May 2026. It approves the bill and sends it to Congress.
  • 12 June 2026. It is published in the Congress Bulletin (file 121/000096).
  • 6 October 2026. It appears as lapsed in Congress: it fell with the dissolution of the Cortes ahead of the elections of 29 November. To pass it, it would have to be submitted and processed again.

That bill designated AESIA as the default supervisory authority, the AEPD for border management and certain biometric uses, and the CGPJ for the justice system. Today it is not applicable law. How and by whom you would be penalised in Spain is still to be defined. The Regulation's obligations, however, already apply. More detail in what penalties an SME can receive.

Where do I start?

  1. Make an inventory. Note which AI tools are used in your company, who uses them and what for. Including the ones nobody approved.
  2. Classify each use with the table above.
  3. Train your team and write some rules of use. Our security guide helps.
  4. Review your chatbots. They should introduce themselves as AI from the first message.
  5. If you use AI in recruitment or credit, prepare human oversight, logs and the notice to staff before 2 December 2027.

If you are going to commission an AI system and want it compliant by design, tell us about it through contact or see how we work with AI agents. And if the cost of putting it in place worries you, take a look at the Artificial Intelligence Voucher (Bono de inteligencia artificial).

This guide is for general information and is up to date as of 25 September 2026; the status of the Spanish law, as of 4 November 2026. It is not legal advice. For a specific case, consult a professional.

Found it useful? Share it with someone who needs it

Xora IntelligenceAI consultancy and custom software in Los Barrios (Cádiz). We write about what we build for businesses in the province. If you have a question about your case, write to us.

Frequently asked questions

Can't see yours? Ask us

Does the EU AI Act apply to an SME that only uses ChatGPT or a chatbot?

Yes, though with little weight. If you use AI in your professional activity you are a deployer. You have to promote AI training for your team (Article 4) and, if you put a chatbot in front of customers or publish deepfakes, meet the transparency obligations of Article 50, which apply from 2 August 2026.

Was the AI Act delayed by the Digital Omnibus?

In part. Regulation (EU) 2026/1744 entered into force on 27 July 2026 and postponed the high-risk obligations: to 2 December 2027 for Annex III (employment, credit, education) and to 2 August 2028 for regulated products. Transparency and the prohibitions were not delayed.

Is using AI to filter CVs high risk?

Yes. Annex III of the Regulation includes AI systems for analysing and filtering job applications and evaluating candidates. As a company that uses them, your high-risk obligations apply from 2 December 2027: use it according to the instructions, human oversight, keep logs and tell staff and their representatives beforehand.

Do I have to say that my chatbot is an AI?

Yes, unless it is obvious. Article 50 requires that people know they are talking to an AI system, clearly and at the latest at the first interaction. Formally the obligation falls on the system's provider, but if it was built for you and you launch it under your name, you are the provider.

How much is the fine for not complying with the AI Act?

It depends on the infringement. Using a prohibited practice: up to €35 million or 7% of worldwide turnover. Breaching transparency or high-risk obligations as a user: up to €15 million or 3%. For SMEs the lower of the two figures applies. In Spain, the bill that was going to set out the procedure lapsed in October 2026.

Who enforces the AI Act in Spain?

The Spanish Agency for the Supervision of Artificial Intelligence (AESIA), based in A Coruña, is the intended authority. The organic law bill that set out authorities and the penalty procedure lapsed on 6 October 2026 with the dissolution of the Cortes, so that part is still pending.

Keep readingOther guides

All guides

Bay of Gibraltar · photo by Cabeza2000 (CC BY-SA 4.0)

Shall we talkabout your business?

Tell us which tasks take up most of your time. We will tell you whether they can be automated and how we would do it. If we can't help, we will tell you that too.