The EU AI Act is the EU law on artificial intelligence. It sorts AI systems by risk, bans a few uses, sets strict requirements for high risk and requires anyone who uses AI in their business to make sure staff have sufficient AI literacy. Most companies are users of AI systems. For them the work starts with knowing what is in use, telling people when they meet an AI, letting a person review what matters and writing down how it works. This text is general information, and a lawyer should assess your individual case.
What the AI Act is
The AI Act is Regulation (EU) 2024/1689. The European Commission describes it as the first comprehensive legal framework on AI, and it rests on risk: the greater the risk a use poses to people, the stricter the requirements.
The rules address two groups. Providers develop AI systems or place them on the market. Deployers use an AI system in their business. A company that takes a ready-made AI service and lets it work on email, documents or cases is usually a deployer, and that is the role this guide assumes.
The four risk levels
The Commission sorts AI use into four levels. The level decides which requirements apply.
- Unacceptable risk. Banned. Examples are social scoring and emotion recognition in workplaces and education. The Commission page lists nine banned practices.
- High risk. Strict requirements before a system is put to use. This covers, for example, AI used in recruitment or to assess employees. Anyone who uses a high-risk system must, among other things, make sure people oversee it and report serious incidents.
- Transparency risk. Information duties. People must be told when they talk to a chatbot, and AI-generated content must be labelled where required, for example deepfakes.
- Minimal risk. No specific rules. The Commission gives AI in video games and spam filters as examples.
What a company should check
Six questions give a good first picture. A few people can answer them in an afternoon.
- What do you use? Write a list of AI services and agents, who owns each one and which tasks it works on.
- Is anything banned? Compare the list with the banned uses. It is rare in an ordinary company, and it is good to have checked.
- Does anything touch people's rights? AI that affects hiring, the assessment of employees or access to services may count as high risk. Ask a lawyer before you go further.
- Do people know they meet an AI? Say so when a customer or colleague talks to an AI, and label content that AI created where required.
- Does a person review what matters? Decide where a person reads the result before anything is sent or changed, and write it down.
- Are the agreements in place? Check data processing agreements and what the provider does with your data.
The AI literacy requirement
According to the European Commission, the AI literacy requirements have applied since 2 February 2025. The requirement is in Article 4 of the regulation: those who use AI systems must make sure staff have sufficient knowledge for their role.
In practice it is enough that staff know what the tool can and cannot do, how data is handled and when a person should review. Short sessions together on real work function well, and our guide on how to train a leadership team in AI describes a setup.
Dates and supervision
According to the European Commission page, the regulation entered into force on 1 August 2024. The banned uses and the AI literacy requirements began to apply on 2 February 2025, and the rules on governance and on general-purpose AI models on 2 August 2025.
The rest applies in stages, and the timing of the high-risk rules has been moved in a later amendment. Check the current dates on the Commission page before you plan around them.
Each EU country must designate national authorities that check compliance. In Sweden the work of assigning the roles has been under way, and the situation changes. Follow regeringen.se for what applies when you read this.
If you use AI agents
An agent that reads incoming email, compiles a week or prepares a draft usually falls at the transparency or minimal risk level. The assessment changes if the agent is used to make or prepare decisions that concern people, for example in recruitment. Then you need to ask a lawyer and plan for human review.
This is how we work at Agentverkstad: we sign a data processing agreement before we touch any data, the price is fixed per step with a cap, and there is a stop point after each step where you look at the result. Our page on trust and data protection goes through the details.
General information
This text is general information and does not replace legal advice. The rules change, and the right assessment depends on how your company uses AI. Let a lawyer or your data protection officer assess your case.
Source: the European Commission page on the AI Act (digital-strategy.ec.europa.eu).
Common questions
Does the AI Act apply to a company that only uses AI?
Yes. The rules address those who develop AI and those who use it in their business. Users are called deployers, and among other things the AI literacy requirement and the information duties apply to them.
Which risk levels are there?
Four: unacceptable risk, which is banned, high risk, transparency risk and minimal risk. The level decides which requirements apply.
What is AI literacy under the regulation?
That staff who work with AI have sufficient knowledge for their role. According to the Commission the requirement has applied since 2 February 2025.
Is an AI agent high risk?
It depends on what it is used for. An agent that prepares email drafts is one thing and an agent that assesses job applicants is another. Let a lawyer assess uses that touch people's rights.
Which authority checks compliance in Sweden?
Each EU country must designate national authorities. How the roles are assigned in Sweden has been under way, so check regeringen.se for the current situation.
Can Agentverkstad decide whether our use is allowed?
No. We build agents and map the work, and we raise the question early so you know when a legal assessment is needed. The assessment itself is made by a lawyer.