Skip to content

The EU AI Act for SMEs: What applies now and what to do

The AI Act explained: roles, risk levels, AI literacy, transparency, deadlines after the Digital Omnibus and four first steps for your business.

Marc Herzog Published on 6 min read

The EU’s AI regulation, the AI Act, already applies in part. For most small and mid-sized businesses, the obligations are manageable and easy to put into practice. This overview shows what applies, what changed in 2026 and where to start.

Key points

  • Prohibitions and the AI literacy obligation have applied since February 2025, transparency obligations since August 2026.
  • If you use AI tools such as chatbots or writing assistants in your business, you are usually a deployer, not a provider.
  • For most SMEs this means: get an overview, set ground rules, train your staff.

What is the AI Act?

The AI Act is Regulation (EU) 2024/1689. It entered into force on 1 August 2024 and applies directly in all EU member states, without needing national implementing legislation.

The rules are risk-based: the more an AI application can affect health, safety or fundamental rights, the stricter the requirements.

In July 2026 the AI Act was amended by the Digital Omnibus Regulation (EU) 2026/1744. Above all, it postponed the deadlines for high-risk AI and changed individual obligations. It has been in force since 27 July 2026.

What is an AI system? Broadly, a machine-based system that infers from its inputs how to generate outputs such as predictions, content, recommendations or decisions (Art. 3(1)). Typical examples are chatbots, text and image generators and AI-supported analyses.

Provider or deployer: what is your role?

Your obligations depend mainly on your role:

Role Who is it? Examples
Provider (Art. 3(3)) Develops an AI system, or has one developed, and places it on the market under its own name. Software companies, AI start-ups, AI model developers
Deployer (Art. 3(4)) Uses an AI system under its own authority, except for purely personal use. A tax advisory firm uses an AI writing assistant, a retailer a chatbot on its website

Most SMEs are deployers and therefore carry far fewer obligations. But note: anyone who markets a high-risk AI system under their own name, substantially modifies it or repurposes an AI system for a high-risk use becomes a provider themselves (Art. 25).

The four risk levels

In simplified terms, the AI Act distinguishes four levels. Typical SME applications sit in the lower two.

Level Examples What applies
Prohibited Social scoring, manipulative techniques, emotion recognition in the workplace prohibited since 2 February 2025
High risk AI in recruitment, creditworthiness checks or exam assessment strict obligations from 2 December 2027
Limited risk Chatbots, deepfakes, AI-generated text and images transparency obligations since 2 August 2026
Minimal risk Spam filters, spell checkers, AI in video games no specific obligations, AI literacy still applies

The classification always depends on the specific purpose, not on the tool alone. The same language model can be harmless in customer service and high-risk in recruitment.

AI literacy: an obligation with room to manoeuvre

Since 2 February 2025, Art. 4 requires providers and deployers to take measures to support the AI literacy of their staff, appropriate to their prior knowledge, the context of use and the people affected.

Since 27 July 2026, Art. 4 has been worded more softly: a specific level of literacy no longer has to be guaranteed. The Commission and member states are to support businesses, especially SMEs.

In practice this means:

  • No certificate is required. Training can be internal or external.
  • Tailor the content to roles and the tools in use.
  • Documenting your measures is not explicitly required but recommended, so you can demonstrate them.
  • Refresh knowledge when tools or uses change.

Transparency: AI must be recognisable

The transparency obligations of Art. 50 have applied since 2 August 2026. For SMEs the key points are:

  • Chatbots: people must be told they are interacting with an AI unless it is obvious. This is a provider obligation, but if you develop a chatbot yourself or offer it under your own name, you may be the provider.
  • Deepfakes: anyone who creates deceptively realistic images, video or audio with AI must disclose it. There are exemptions for art and satire.
  • AI text for the public: text on matters of public interest must be labelled unless it has been editorially reviewed and someone takes responsibility for it.

The notice must be given clearly and accessibly at the latest at the first interaction.

Prohibited AI practices

Since February 2025, eight practices have been prohibited, including manipulation through subliminal techniques, social scoring and emotion recognition in the workplace and in education. The last point is particularly relevant for SMEs, for example in call-centre or HR tools.

From 2 December 2026, two more prohibitions apply: AI systems that create intimate deepfakes of real people without their consent, and systems that generate child sexual abuse material.

Data protection: the GDPR still applies

The AI Act does not replace the GDPR. As soon as personal data flows into an AI tool, both apply. In practice, data protection is often the first issue to address. Typical checkpoints are the legal basis, a data processing agreement with the tool provider, transfers outside the EU and informing the people concerned. Involve your data protection officer early.

Key deadlines

Date What applies
1 August 2024 AI Act enters into force
2 February 2025 Prohibitions and AI literacy
2 August 2025 Rules for general-purpose AI models, penalties
27 July 2026 Digital Omnibus in force, deadlines postponed
2 August 2026 General application: transparency, supervision
2 December 2026 Two new prohibitions, end of transition for technical AI labelling
2 December 2027 High-risk AI under Annex III, e.g. HR and credit
2 August 2028 High-risk AI in products (Annex I)

Supervision and fines

In Germany, the Federal Network Agency (Bundesnetzagentur) is the central market surveillance authority and point of contact, based on the AI Market Surveillance Act in force since 29 July 2026. Other authorities are responsible for certain sectors. Its AI Service Desk offers information for SMEs and an online risk-assessment tool (in German).

Fines under Art. 99 reach up to €35 million or 7% of worldwide annual turnover for prohibited practices. For SMEs, the lower of the two amounts applies. Art. 99 sets no separate fine range for AI literacy under Art. 4.

Four first steps for your business

Regardless of the risk level of your tools, these four steps create a solid foundation:

  1. AI inventory: record which AI tools are used, including via personal accounts, what for and with which data.
  2. Ground rules: set out in an AI policy which tools are allowed, which data is off-limits and how results are checked. You will find a template in the article AI policy for your team.
  3. Training: enable staff according to their role and record who learned what and when.
  4. Responsibility: name a contact person who keeps track and reviews new tools.

Also worthwhile: check the labelling of chatbots and AI content, assign tools to a risk level and involve data protection.

If you would like to tackle this with your team: in my AI literacy workshop we spend half a day on the fundamentals, the AI Act, safe handling of data and a draft of your AI ground rules.


Sources: Regulation (EU) 2024/1689 · Regulation (EU) 2026/1744 · European Commission: AI literacy Q&A · Bundesnetzagentur: AI Service Desk. As of November 2026. Simplified presentation; the text of the regulation is authoritative.

This article provides general information and is not legal advice. For binding advice on your specific case, please consult a lawyer.

Let’s find out whether AI pays off for you.

In the free initial call, you describe your situation. You get an initial assessment and a suggestion for the next step.

Prefer to talk directly? +49 160 3759917 or kontakt@marc-herzog-consulting.de