EU AI Act: What small businesses really need to do now
The EU AI Act explained simply: which obligations apply to small businesses, which deadlines the Digital Omnibus postponed in 2026, and a 5-step checklist.

Whether it's ChatGPT for emails, Copilot in Office or a chatbot on the website: artificial intelligence has long been part of everyday life in many small businesses. At the same time, the EU AI Act (German: "KI-Verordnung") has been in force since August 2024. Many business owners are asking: Does this even affect me? And what exactly do I have to do?
The good news first: most of the strict obligations apply to makers of risky AI systems, not to the trade business that uses AI to write quotes. A few obligations do apply to every company that uses AI, though. And in 2026 the EU changed the timeline once again. Here is where things stand, explained in plain language.
Important: This article gives an overview and is not a substitute for legal advice. As of: October 2026.
In short
- If you use AI in your business, you count as a deployer and have a few manageable obligations.
- Since February 2025 you have to make sure your team builds AI literacy.
- Since August 2026 transparency obligations apply, for example for chatbots and deceptively realistic AI images.
- The strict rules for high-risk AI (for example in selecting job applicants) have been postponed to December 2027.
- In Germany, the Bundesnetzagentur (Federal Network Agency) is the central supervisory authority.
Am I affected at all?
The regulation mainly distinguishes between two roles:
- Providers develop an AI system or bring it to market under their own name, for example a software manufacturer.
- Deployers use an AI system in their company, for example a painting business that drafts quote texts with ChatGPT.
So almost every small company that uses AI is a deployer. Only purely private use is exempt. You need to be careful if you offer an AI system under your own name, such as a chatbot you built yourself: then you may also count as a provider.
The four risk levels, simply explained
The regulation sorts AI by its risk. The higher the risk, the stricter the rules.
| Level | Examples | What applies |
|---|---|---|
| Unacceptable risk | Social scoring, emotion recognition in the workplace | Prohibited |
| High risk | AI in selecting job applicants, creditworthiness, education | Strict obligations |
| Limited risk | Chatbots, AI-generated images and videos | Transparency obligations |
| Minimal risk | Spam filters, text drafts, translations | No special obligations |
Most applications in small companies fall into the bottom two levels.
These deadlines apply
With the so-called Digital Omnibus (Regulation (EU) 2026/1744), the EU postponed some deadlines in 2026. This is what the timeline looks like now:
| Date | What applies |
|---|---|
| 1 August 2024 | The EU AI Act enters into force |
| 2 February 2025 | Bans on certain AI practices and the AI literacy obligation |
| 2 August 2025 | Obligations for providers of large AI models (such as GPT or Claude) |
| 2 August 2026 | Transparency obligations under Art. 50, start of supervision |
| 2 December 2026 | New bans (e.g. non-consensual intimate deepfakes) |
| 2 December 2027 | Obligations for high-risk AI under Annex III (e.g. personnel selection) |
| 2 August 2028 | Obligations for high-risk AI in regulated products (Annex I) |
Obligation 1: AI literacy in your team (Art. 4)
Since February 2025, companies that use AI have to take steps to make sure their employees use AI competently. The Digital Omnibus softened this obligation somewhat: you have to take measures to promote AI literacy. But you don't have to guarantee a specific level of knowledge for every single person.
Here's how to put it into practice:
- a short training session: What can AI do, where does it make mistakes, what must not go into it?
- a simple AI policy with approved tools and clear rules, such as "no customer data in free AI tools"
- briefly document both: who was trained and when
Tip: A short training session and a one-page policy are a good start. What is appropriate for your business depends on how you use AI.
Obligation 2: Transparency (Art. 50)
Since 2 August 2026 the rule is: people should know when they are dealing with AI.
- Chatbots: Anyone chatting with an AI chatbot must be able to tell that it is an AI, unless this is obvious. This obligation falls on the provider. If you run a chatbot under your name, you should definitely implement it anyway. A clear notice such as "I am an AI assistant" is enough.
- Deepfakes: You must label deceptively realistic AI images, videos or voices of real people, places or events as AI-generated.
- Texts on public matters: AI texts intended to inform the public about matters of public interest must be labelled, unless a human has reviewed them and takes editorial responsibility for them.
There is no labelling obligation for normal advertising copy, emails or quotes that you prepare with AI and check yourself. Nor do you have to label every AI image: the obligation for deployers concerns deepfakes, meaning content that resembles real people, places or events and falsely appears genuine.
Obligation 3: Avoid prohibited practices
Since February 2025 some applications have been banned outright. Two are especially relevant for small companies:
- Emotion recognition in the workplace, for example software that uses a camera to analyse employees' mood (exceptions exist only for medical or safety reasons).
- Manipulative techniques that push people unconsciously into harmful decisions or exploit their weaknesses.
If you use AI for texts, images or analyses, this does not affect you.
Special case: AI in hiring and HR
This is where it gets serious. AI that filters applications, rates candidates or helps decide on promotions and dismissals counts as high-risk AI. From 2 December 2027, deployers will then have to, among other things:
- use the system according to the manufacturer's instructions,
- ensure human oversight by trained people,
- monitor operation and keep logs,
- inform employees before such a system is used.
In practice: If you are currently looking into HR software with AI features, get the provider to confirm in writing how it complies with the EU AI Act.
Who enforces this in Germany?
Germany has passed the KI-Marktüberwachungsgesetz (KI-MIG) (German AI Market Surveillance Act) for this purpose. It has been in force since the end of July 2026. The central supervisory authority is the Bundesnetzagentur. In some areas, specialist authorities remain responsible, such as BaFin (Germany's Federal Financial Supervisory Authority) for banks and insurers. The Bundesnetzagentur is also the point of contact for complaints.
What penalties can you face?
The fines are high, but they are mainly designed for major violations:
- up to 35 million euros or 7% of worldwide annual turnover for prohibited practices,
- up to 15 million euros or 3% for violations of other obligations, such as transparency,
- up to 7.5 million euros or 1% for supplying incorrect information to authorities.
For small and medium-sized enterprises, the lower of the two amounts applies in each case. The regulation itself does not provide for a separate fine for a lack of AI literacy. But if damage occurs, missing training can count against you in liability questions.
Checklist: 5 steps to get started
- Create an AI inventory: Which AI tools are used in the business, by whom, and for what?
- Clarify your role: Are you only a deployer, or do you offer something under your own name?
- Assess the risk: Is anything prohibited or high-risk involved, for example in HR?
- Build AI literacy: a short training session, a one-page AI policy, and keep data protection in mind (GDPR, data processing agreement).
- Implement transparency and document it: chatbot notice, labelling of deepfakes, review once a year.
Conclusion
The EU AI Act is no reason to give up on AI. For most small companies it means: train your team, write down the rules, label honestly. For most businesses that is manageable, and it builds trust with customers along the way.
If you're only planning to use AI, first read where AI helps in a business and where its limits are.
How we can help
In our Artificial intelligence area, we help you draw up an overview of your AI tools, write a simple AI policy and add the notice to your chatbot. This does not replace legal advice. For individual questions, you need a lawyer.
In a free initial consultation, we look together at which AI tools you use and which of them should be dealt with first.
Sources & further reading
- Verordnung (EU) 2024/1689 über künstliche Intelligenz (EUR-Lex) eur-lex.europa.eu
- Digital Omnibus on AI, Verordnung (EU) 2026/1744 (AI Act Explorer) artificialintelligenceact.eu
- Gesetz zur Marktüberwachung und Innovationsförderung von KI (KI-MIG) gesetze-im-internet.de
- Deutscher Bundestag: Ja zur Durchführung der Verordnung über künstliche Intelligenz bundestag.de


