What Is the AI Act? What Changes for You and Your Work
The AI Act is the EU law for AI. Find out what already applies, what has been delayed until 2027 and 2028, and what you need to do.

What is the AI Act if you use ChatGPT, Copilot or Gemini at work? The AI Act is the European law that sets rules for AI, mainly based on risk. Some rules already apply, while the toughest rules for high-risk AI have been delayed by the Digital Omnibus until 2027 and 2028.
That sounds legal, but the basics are easy to understand. The law does not only affect large AI companies. It also affects organisations that use AI in their work.
What is the AI Act, and who does it apply to?
The AI Act is a European regulation that sets rules for developing and using AI. In Dutch, this law is also called the AI Regulation.
Because it is a European regulation, the law applies directly in all EU countries. So it applies in the Netherlands and Belgium too. The law is not about the language you use AI in, but about how you use AI.
According to the European Commission, the AI Act works with obligations for different roles in the AI chain. Think of providers that build or supply AI systems, but also organisations that use AI in their own work. The European Commission explains this on its page about the regulatory framework for AI.
A provider is, for example, a party that places an AI system or AI model on the market. A user is an organisation that uses such a system. In the law, that user is often called a deployer.
A company that uses Copilot, a school that uses AI or a sole trader who uses ChatGPT for client work can therefore be a user in the meaning of the law. If you are a private user, most of the obligations do not apply to you. If you only use AI at home to make a recipe or plan a holiday, the law is usually not aimed at you.
If you use AI for work, it becomes more important. You do not need to know the whole law by heart. The core point is simple: the greater the risk to people, the stricter the rules.
How does the AI Act use risk levels?
The AI Act uses four risk levels: unacceptable risk, high risk, limited risk and minimal risk. The European Commission describes this risk-based approach on its page about the AI regulatory framework.
This helps to place ordinary AI tools better. Not every AI tool is dangerous or tightly regulated. But some uses can have major consequences for people, for example if AI helps with selection, assessment or access to important services.
| Level | What it means | Example | What is required |
|---|---|---|---|
| Unacceptable risk | This kind of AI may not be used. | Social scoring. | Do not use it. |
| High risk | AI can have major consequences for someone’s opportunities, rights or safety. | AI used to assess job applicants or creditworthiness. | Strict requirements, extra care and human oversight. |
| Limited risk | People must know they are dealing with AI. | A chatbot that talks to customers. | Make it clear that it is AI. |
| Minimal risk | Many ordinary uses fall here. | A spam filter. | Almost free to use, with common sense. |
The difference is therefore not only in the technology. It is mainly about what the AI does and what effect that has on people. A writing aid for an email is something else entirely from AI that ranks job applicants.
With limited risk, transparency is often the key issue. That means you have to be clear about the use of AI. For some AI content, it can also be important that people know something was made with AI. That matters, for example, if you want to understand what a deepfake is.
So do not label an application as high risk too quickly. First look at the purpose of the tool. If you use AI to draft an email, that is usually different from AI that helps decide whether someone gets hired.
Which AI Act rules already apply?
Part of the AI Act already applies. According to the European timetable on the European Commission page, the AI Act has been in force since August 2024 and is being introduced in stages. Banned practices and AI literacy have applied since February 2025, and rules for general-purpose AI models since August 2025.
Banned practices are uses that the law sees as posing an unacceptable risk. Think of examples such as social scoring. Such uses are not normal office use of AI.
AI literacy means people know enough to use AI responsibly. It is not about programming. It is about understanding what AI can and cannot do, where the risks are, and when you should have a person check the result.
For workers, this is immediately practical. If you use AI for texts, summaries, customer contact or internal plans, you need to know that AI can make mistakes. You also need to know that you should not simply put sensitive data into a public tool.
For entrepreneurs, AI literacy is something you need to organise. You can start small with clear rules and a short explanation for your team. For that, also read the explanation of AI literacy for entrepreneurs.
The rules for general-purpose AI models mainly affect the providers of those models. Think of parties that build broad AI models and make them available. Still, you may notice the effects as a user, for example in the information providers give or in limits that tools receive.
What matters is that these rules have not disappeared because of the later delay. Banned practices remain banned. Making sure people have enough AI knowledge remains important too.
What changes because of the delay to the AI Act?
Because of the Digital Omnibus, mainly the toughest rules for high-risk AI have been delayed. The banned practices and AI literacy do not disappear.
The Digital Omnibus is Regulation (EU) 2026/1744. According to AI Compliance Hub, it was published on 24 July 2026 and entered into force on 27 July 2026. It also states that the political agreement was reached on 7 May 2026.
The main shift is about high-risk AI. According to AI Compliance Hub, high-risk AI in Annex III moves to 2 December 2027. High-risk AI in Annex I moves to 2 August 2028.
DLA Piper explains that the European Commission proposed the delay on 19 November 2025 and that the Omnibus entered into force on 27 July 2026 after its publication on 24 July 2026.
| Moment | What happens |
|---|---|
| August 2024 | The AI Act enters into force in stages. |
| February 2025 | Banned practices and AI literacy apply. |
| August 2025 | Rules for general-purpose AI models apply. |
| 19 November 2025 | The European Commission proposes the delay. |
| 7 May 2026 | A political agreement on the Digital Omnibus is reached. |
| 24 July 2026 | Digital Omnibus published. |
| 27 July 2026 | Digital Omnibus enters into force. |
| 2 December 2027 | New date for high-risk AI from Annex III. |
| 2 August 2028 | New date for high-risk AI from Annex I. |
The delay is therefore not a free pass to do nothing. If you use AI in decisions about people, you need to pay close attention now. Think about recruitment, credit or other choices that affect someone’s opportunities.
Use the extra time to get an overview. Do not wait until the date is close. It becomes harder then to make changes calmly.
Also watch out for old information. Some explanations still mention 2 August 2026 for high-risk AI. After the Digital Omnibus, that is no longer the correct date for the high-risk obligations mentioned here.
What does the AI Act mean for you if you use ChatGPT or Copilot?
For you as an ordinary user at work, the AI Act mainly means that you should use AI carefully. Check AI results, do not put sensitive data into public tools and follow your employer’s rules.
If you use ChatGPT, Copilot or Gemini to write a text, you are usually not dealing with high-risk AI. But you can still make mistakes. AI can give wrong information, make things up or process confidential information in a way that is not allowed.
The GDPR still applies alongside the AI Act. The GDPR is about personal data. The AI Act is about AI systems and the risk of using them.
That means you need to be extra careful with names, customer data, medical information, employee data or other sensitive data. Also read how to use AI safely in practice.
The difference between user and provider remains important. You usually do not build an AI model. But your organisation does use an AI tool. That is why your organisation may need to make agreements about safe and responsible use.
For some AI applications, the law has transparency obligations. Think of situations where people need to know they are talking to AI or that certain content was made with AI. Always check the current guidance from the European Commission for this, especially if you use chatbots, images, audio or video.
What should you arrange now as an entrepreneur for the AI Act?
As an entrepreneur, start with an overview of the AI tools you use. Then agree on rules and make sure your staff know enough to use AI responsibly.
You do not need to produce a thick legal document straight away. A short, clear approach is often enough. Record which tools you use, what you use them for and who checks the results.
This checklist will help you get started:
- Make a list of all the AI tools you use.
- Determine the risk level for each tool.
- Agree that personal data does not go into public tools without good reason.
- Give staff a short AI training session.
- Record who checks AI outputs.
- Put 2 December 2027 and 2 August 2028 in your calendar.
- Check extra carefully when AI helps make decisions about people.
An AI policy does not need to be complicated. It mainly needs to be clear for the people who work with it. Who may use which tool? What data may go in? When does a human need to check the result?
If you want to tackle this practically, start with setting up an AI policy. That way you make agreements before something goes wrong.
Be extra careful with AI that assesses, ranks or selects people. Examples are recruitment, credit, access to training or other decisions that can have major consequences. If in doubt, ask an expert and consult the official text.
Where can you find current information about the AI Act?
Look for current information from official sources and from sources that clearly state which date they use. The supervisory structure and practical implementation can still be filled in differently by each country.
For the Netherlands and Belgium, the same European basis applies, because the AI Act is an EU regulation. Practical implementation can differ from country to country. So be careful with old blogs, old dates and loose lists on social media.
The safest approach is simple. Check the European page about the regulatory framework for AI and compare new information with the dates in the Digital Omnibus.
Practical example: Marieke maps out her AI use
Marieke can assess her risks better by first writing down her AI tools. She is a sole trader, makes client emails with ChatGPT and also uses a tool to sort applications for an intern.
She does not start with legal details. She first writes down which AI tools she uses. Then she puts each tool into a risk level.
- She notes that she uses ChatGPT to write client emails.
- She notes that she wants to use a chatbot on her website.
- She notes that she uses a tool to rank applicants.
Then she looks at the risk. Writing emails is usually minimal risk. A chatbot on her site is limited risk, because visitors need to know they are talking to AI.
The tool that ranks applicants is more sensitive. AI in recruitment can fall under high risk and needs extra care. Marieke therefore records that a person always makes the final decision.
This example is not legal advice. It shows how you can make an initial assessment. If in doubt, consult the official text or a lawyer.
The expected result is a table with tool, level, reason and action. Use that as an initial indication. Do not copy it blindly.
Common mistakes with the AI Act
You avoid a lot of hassle by not making the AI Act bigger or smaller than it is. These mistakes soon show up in the workplace.
- Thinking the AI Act only applies to big tech companies. Users also have obligations, such as making sure staff have AI literacy.
- Thinking everything has been delayed. Only the high-risk obligations have been delayed. The bans and AI literacy are still in place.
- Confusing the AI Act with the GDPR. The GDPR is about personal data. The AI Act is about AI systems and risks.
- Calling every AI application high risk. Many everyday uses are minimal or limited risk.
- Repeating old dates. Check the current dates after the Digital Omnibus.
The solution is usually simple. First distinguish between normal office work with AI and AI that affects decisions about people. The second group needs much more attention.
Frequently asked questions
Does the AI Act also apply to sole traders?
Yes, if you use AI professionally, you can be a user in the meaning of the law. What obligations you have depends on the application.
Is ChatGPT banned by the AI Act?
No, ordinary chatbots are not banned. The AI Act only bans certain practices with unacceptable risk.
Is the AI Act already in force?
Yes, according to the European Commission the AI Act has been in force since August 2024 and is being introduced in stages. Some parts already apply, while other rules come later.
What has been delayed?
The high-risk obligations have been delayed. For high-risk AI from Annex III, the new date is 2 December 2027, and for Annex I it is 2 August 2028.
Do I need to label AI texts?
For certain AI applications, the law has transparency obligations. Always check the current situation, especially for chatbots, deepfakes and content that could mislead people.
The best next step is simple: make an overview of your own AI tools today. Then you can calmly make agreements and set up a simple AI policy.
Was this article helpful?




